birefnet.py

### config.py

import os
import math


class Config():
    def __init__(self) -> None:
        # PATH settings
        self.sys_home_dir = os.path.expanduser('~')     # Make up your file system as: SYS_HOME_DIR/codes/dis/BiRefNet, SYS_HOME_DIR/datasets/dis/xx, SYS_HOME_DIR/weights/xx

        # TASK settings
        self.task = ['DIS5K', 'COD', 'HRSOD', 'DIS5K+HRSOD+HRS10K', 'P3M-10k'][0]
        self.training_set = {
            'DIS5K': ['DIS-TR', 'DIS-TR+DIS-TE1+DIS-TE2+DIS-TE3+DIS-TE4'][0],
            'COD': 'TR-COD10K+TR-CAMO',
            'HRSOD': ['TR-DUTS', 'TR-HRSOD', 'TR-UHRSD', 'TR-DUTS+TR-HRSOD', 'TR-DUTS+TR-UHRSD', 'TR-HRSOD+TR-UHRSD', 'TR-DUTS+TR-HRSOD+TR-UHRSD'][5],
            'DIS5K+HRSOD+HRS10K': 'DIS-TE1+DIS-TE2+DIS-TE3+DIS-TE4+DIS-TR+TE-HRS10K+TE-HRSOD+TE-UHRSD+TR-HRS10K+TR-HRSOD+TR-UHRSD',     # leave DIS-VD for evaluation.
            'P3M-10k': 'TR-P3M-10k',
        }[self.task]
        self.prompt4loc = ['dense', 'sparse'][0]

        # Faster-Training settings
        self.load_all = True
        self.compile = True     # 1. Trigger CPU memory leak in some extend, which is an inherent problem of PyTorch.
                                #   Machines with > 70GB CPU memory can run the whole training on DIS5K with default setting.
                                # 2. Higher PyTorch version may fix it: https://github.com/pytorch/pytorch/issues/119607.
                                # 3. But compile in Pytorch > 2.0.1 seems to bring no acceleration for training.
        self.precisionHigh = True

        # MODEL settings
        self.ms_supervision = True
        self.out_ref = self.ms_supervision and True
        self.dec_ipt = True
        self.dec_ipt_split = True
        self.cxt_num = [0, 3][1]    # multi-scale skip connections from encoder
        self.mul_scl_ipt = ['', 'add', 'cat'][2]
        self.dec_att = ['', 'ASPP', 'ASPPDeformable'][2]
        self.squeeze_block = ['', 'BasicDecBlk_x1', 'ResBlk_x4', 'ASPP_x3', 'ASPPDeformable_x3'][1]
        self.dec_blk = ['BasicDecBlk', 'ResBlk', 'HierarAttDecBlk'][0]

        # TRAINING settings
        self.batch_size = 4
        self.IoU_finetune_last_epochs = [
            0,
            {
                'DIS5K': -50,
                'COD': -20,
                'HRSOD': -20,
                'DIS5K+HRSOD+HRS10K': -20,
                'P3M-10k': -20,
            }[self.task]
        ][1]    # choose 0 to skip
        self.lr = (1e-4 if 'DIS5K' in self.task else 1e-5) * math.sqrt(self.batch_size / 4)     # DIS needs high lr to converge faster. Adapt the lr linearly
        self.size = 1024
        self.num_workers = max(4, self.batch_size)          # will be decrease to min(it, batch_size) at the initialization of the data_loader

        # Backbone settings
        self.bb = [
            'vgg16', 'vgg16bn', 'resnet50',         # 0, 1, 2
            'swin_v1_t', 'swin_v1_s',               # 3, 4
            'swin_v1_b', 'swin_v1_l',               # 5-bs9, 6-bs4
            'pvt_v2_b0', 'pvt_v2_b1',               # 7, 8
            'pvt_v2_b2', 'pvt_v2_b5',               # 9-bs10, 10-bs5
        ][6]
        self.lateral_channels_in_collection = {
            'vgg16': [512, 256, 128, 64], 'vgg16bn': [512, 256, 128, 64], 'resnet50': [1024, 512, 256, 64],
            'pvt_v2_b2': [512, 320, 128, 64], 'pvt_v2_b5': [512, 320, 128, 64],
            'swin_v1_b': [1024, 512, 256, 128], 'swin_v1_l': [1536, 768, 384, 192],
            'swin_v1_t': [768, 384, 192, 96], 'swin_v1_s': [768, 384, 192, 96],
            'pvt_v2_b0': [256, 160, 64, 32], 'pvt_v2_b1': [512, 320, 128, 64],
        }[self.bb]
        if self.mul_scl_ipt == 'cat':
            self.lateral_channels_in_collection = [channel * 2 for channel in self.lateral_channels_in_collection]
        self.cxt = self.lateral_channels_in_collection[1:][::-1][-self.cxt_num:] if self.cxt_num else []

        # MODEL settings - inactive
        self.lat_blk = ['BasicLatBlk'][0]
        self.dec_channels_inter = ['fixed', 'adap'][0]
        self.refine = ['', 'itself', 'RefUNet', 'Refiner', 'RefinerPVTInChannels4'][0]
        self.progressive_ref = self.refine and True
        self.ender = self.progressive_ref and False
        self.scale = self.progressive_ref and 2
        self.auxiliary_classification = False       # Only for DIS5K, where class labels are saved in `dataset.py`.
        self.refine_iteration = 1
        self.freeze_bb = False
        self.model = [
            'BiRefNet',
        ][0]
        if self.dec_blk == 'HierarAttDecBlk':
            self.batch_size = 2 ** [0, 1, 2, 3, 4][2]

        # TRAINING settings - inactive
        self.preproc_methods = ['flip', 'enhance', 'rotate', 'pepper', 'crop'][:4]
        self.optimizer = ['Adam', 'AdamW'][1]
        self.lr_decay_epochs = [1e5]    # Set to negative N to decay the lr in the last N-th epoch.
        self.lr_decay_rate = 0.5
        # Loss
        self.lambdas_pix_last = {
            # not 0 means opening this loss
            # original rate -- 1 : 30 : 1.5 : 0.2, bce x 30
            'bce': 30 * 1,          # high performance
            'iou': 0.5 * 1,         # 0 / 255
            'iou_patch': 0.5 * 0,   # 0 / 255, win_size = (64, 64)
            'mse': 150 * 0,         # can smooth the saliency map
            'triplet': 3 * 0,
            'reg': 100 * 0,
            'ssim': 10 * 1,          # help contours,
            'cnt': 5 * 0,          # help contours
            'structure': 5 * 0,    # structure loss from codes of MVANet. A little improvement on DIS-TE[1,2,3], a bit more decrease on DIS-TE4.
        }
        self.lambdas_cls = {
            'ce': 5.0
        }
        # Adv
        self.lambda_adv_g = 10. * 0        # turn to 0 to avoid adv training
        self.lambda_adv_d = 3. * (self.lambda_adv_g > 0)

        # PATH settings - inactive
        self.data_root_dir = os.path.join(self.sys_home_dir, 'datasets/dis')
        self.weights_root_dir = os.path.join(self.sys_home_dir, 'weights')
        self.weights = {
            'pvt_v2_b2': os.path.join(self.weights_root_dir, 'pvt_v2_b2.pth'),
            'pvt_v2_b5': os.path.join(self.weights_root_dir, ['pvt_v2_b5.pth', 'pvt_v2_b5_22k.pth'][0]),
            'swin_v1_b': os.path.join(self.weights_root_dir, ['swin_base_patch4_window12_384_22kto1k.pth', 'swin_base_patch4_window12_384_22k.pth'][0]),
            'swin_v1_l': os.path.join(self.weights_root_dir, ['swin_large_patch4_window12_384_22kto1k.pth', 'swin_large_patch4_window12_384_22k.pth'][0]),
            'swin_v1_t': os.path.join(self.weights_root_dir, ['swin_tiny_patch4_window7_224_22kto1k_finetune.pth'][0]),
            'swin_v1_s': os.path.join(self.weights_root_dir, ['swin_small_patch4_window7_224_22kto1k_finetune.pth'][0]),
            'pvt_v2_b0': os.path.join(self.weights_root_dir, ['pvt_v2_b0.pth'][0]),
            'pvt_v2_b1': os.path.join(self.weights_root_dir, ['pvt_v2_b1.pth'][0]),
        }

        # Callbacks - inactive
        self.verbose_eval = True
        self.only_S_MAE = False
        self.use_fp16 = False   # Bugs. It may cause nan in training.
        self.SDPA_enabled = False    # Bugs. Slower and errors occur in multi-GPUs

        # others
        self.device = [0, 'cpu'][0]     # .to(0) == .to('cuda:0')

        self.batch_size_valid = 1
        self.rand_seed = 7
        # run_sh_file = [f for f in os.listdir('.') if 'train.sh' == f] + [os.path.join('..', f) for f in os.listdir('..') if 'train.sh' == f]
        # with open(run_sh_file[0], 'r') as f:
        #     lines = f.readlines()
        #     self.save_last = int([l.strip() for l in lines if '"{}")'.format(self.task) in l and 'val_last=' in l][0].split('val_last=')[-1].split()[0])
        #     self.save_step = int([l.strip() for l in lines if '"{}")'.format(self.task) in l and 'step=' in l][0].split('step=')[-1].split()[0])
        # self.val_step = [0, self.save_step][0]

    def print_task(self) -> None:
        # Return task for choosing settings in shell scripts.
        print(self.task)



### models/backbones/pvt_v2.py

import torch
import torch.nn as nn
from functools import partial

from timm.models.layers import DropPath, to_2tuple, trunc_normal_
from timm.models.registry import register_model

import math

# from config import Config

# config = Config()

class Mlp(nn.Module):
    def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.):
        super().__init__()
        out_features = out_features or in_features
        hidden_features = hidden_features or in_features
        self.fc1 = nn.Linear(in_features, hidden_features)
        self.dwconv = DWConv(hidden_features)
        self.act = act_layer()
        self.fc2 = nn.Linear(hidden_features, out_features)
        self.drop = nn.Dropout(drop)

        self.apply(self._init_weights)

    def _init_weights(self, m):
        if isinstance(m, nn.Linear):
            trunc_normal_(m.weight, std=.02)
            if isinstance(m, nn.Linear) and m.bias is not None:
                nn.init.constant_(m.bias, 0)
        elif isinstance(m, nn.LayerNorm):
            nn.init.constant_(m.bias, 0)
            nn.init.constant_(m.weight, 1.0)
        elif isinstance(m, nn.Conv2d):
            fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
            fan_out //= m.groups
            m.weight.data.normal_(0, math.sqrt(2.0 / fan_out))
            if m.bias is not None:
                m.bias.data.zero_()

    def forward(self, x, H, W):
        x = self.fc1(x)
        x = self.dwconv(x, H, W)
        x = self.act(x)
        x = self.drop(x)
        x = self.fc2(x)
        x = self.drop(x)
        return x


class Attention(nn.Module):
    def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0., sr_ratio=1):
        super().__init__()
        assert dim % num_heads == 0, f"dim {dim} should be divided by num_heads {num_heads}."

        self.dim = dim
        self.num_heads = num_heads
        head_dim = dim // num_heads
        self.scale = qk_scale or head_dim ** -0.5

        self.q = nn.Linear(dim, dim, bias=qkv_bias)
        self.kv = nn.Linear(dim, dim * 2, bias=qkv_bias)
        self.attn_drop_prob = attn_drop
        self.attn_drop = nn.Dropout(attn_drop)
        self.proj = nn.Linear(dim, dim)
        self.proj_drop = nn.Dropout(proj_drop)

        self.sr_ratio = sr_ratio
        if sr_ratio > 1:
            self.sr = nn.Conv2d(dim, dim, kernel_size=sr_ratio, stride=sr_ratio)
            self.norm = nn.LayerNorm(dim)

        self.apply(self._init_weights)

    def _init_weights(self, m):
        if isinstance(m, nn.Linear):
            trunc_normal_(m.weight, std=.02)
            if isinstance(m, nn.Linear) and m.bias is not None:
                nn.init.constant_(m.bias, 0)
        elif isinstance(m, nn.LayerNorm):
            nn.init.constant_(m.bias, 0)
            nn.init.constant_(m.weight, 1.0)
        elif isinstance(m, nn.Conv2d):
            fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
            fan_out //= m.groups
            m.weight.data.normal_(0, math.sqrt(2.0 / fan_out))
            if m.bias is not None:
                m.bias.data.zero_()

    def forward(self, x, H, W):
        B, N, C = x.shape
        q = self.q(x).reshape(B, N, self.num_heads, C // self.num_heads).permute(0, 2, 1, 3)

        if self.sr_ratio > 1:
            x_ = x.permute(0, 2, 1).reshape(B, C, H, W)
            x_ = self.sr(x_).reshape(B, C, -1).permute(0, 2, 1)
            x_ = self.norm(x_)
            kv = self.kv(x_).reshape(B, -1, 2, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
        else:
            kv = self.kv(x).reshape(B, -1, 2, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
        k, v = kv[0], kv[1]

        if config.SDPA_enabled:
            x = torch.nn.functional.scaled_dot_product_attention(
                q, k, v,
                attn_mask=None, dropout_p=self.attn_drop_prob, is_causal=False
            ).transpose(1, 2).reshape(B, N, C)
        else:
            attn = (q @ k.transpose(-2, -1)) * self.scale
            attn = attn.softmax(dim=-1)
            attn = self.attn_drop(attn)

            x = (attn @ v).transpose(1, 2).reshape(B, N, C)
        x = self.proj(x)
        x = self.proj_drop(x)

        return x


class Block(nn.Module):

    def __init__(self, dim, num_heads, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop=0., attn_drop=0.,
                 drop_path=0., act_layer=nn.GELU, norm_layer=nn.LayerNorm, sr_ratio=1):
        super().__init__()
        self.norm1 = norm_layer(dim)
        self.attn = Attention(
            dim,
            num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale,
            attn_drop=attn_drop, proj_drop=drop, sr_ratio=sr_ratio)
        # NOTE: drop path for stochastic depth, we shall see if this is better than dropout here
        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
        self.norm2 = norm_layer(dim)
        mlp_hidden_dim = int(dim * mlp_ratio)
        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)

        self.apply(self._init_weights)

    def _init_weights(self, m):
        if isinstance(m, nn.Linear):
            trunc_normal_(m.weight, std=.02)
            if isinstance(m, nn.Linear) and m.bias is not None:
                nn.init.constant_(m.bias, 0)
        elif isinstance(m, nn.LayerNorm):
            nn.init.constant_(m.bias, 0)
            nn.init.constant_(m.weight, 1.0)
        elif isinstance(m, nn.Conv2d):
            fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
            fan_out //= m.groups
            m.weight.data.normal_(0, math.sqrt(2.0 / fan_out))
            if m.bias is not None:
                m.bias.data.zero_()

    def forward(self, x, H, W):
        x = x + self.drop_path(self.attn(self.norm1(x), H, W))
        x = x + self.drop_path(self.mlp(self.norm2(x), H, W))

        return x


class OverlapPatchEmbed(nn.Module):
    """ Image to Patch Embedding
    """

    def __init__(self, img_size=224, patch_size=7, stride=4, in_channels=3, embed_dim=768):
        super().__init__()
        img_size = to_2tuple(img_size)
        patch_size = to_2tuple(patch_size)

        self.img_size = img_size
        self.patch_size = patch_size
        self.H, self.W = img_size[0] // patch_size[0], img_size[1] // patch_size[1]
        self.num_patches = self.H * self.W
        self.proj = nn.Conv2d(in_channels, embed_dim, kernel_size=patch_size, stride=stride,
                              padding=(patch_size[0] // 2, patch_size[1] // 2))
        self.norm = nn.LayerNorm(embed_dim)

        self.apply(self._init_weights)

    def _init_weights(self, m):
        if isinstance(m, nn.Linear):
            trunc_normal_(m.weight, std=.02)
            if isinstance(m, nn.Linear) and m.bias is not None:
                nn.init.constant_(m.bias, 0)
        elif isinstance(m, nn.LayerNorm):
            nn.init.constant_(m.bias, 0)
            nn.init.constant_(m.weight, 1.0)
        elif isinstance(m, nn.Conv2d):
            fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
            fan_out //= m.groups
            m.weight.data.normal_(0, math.sqrt(2.0 / fan_out))
            if m.bias is not None:
                m.bias.data.zero_()

    def forward(self, x):
        x = self.proj(x)
        _, _, H, W = x.shape
        x = x.flatten(2).transpose(1, 2)
        x = self.norm(x)

        return x, H, W


class PyramidVisionTransformerImpr(nn.Module):
    def __init__(self, img_size=224, patch_size=16, in_channels=3, num_classes=1000, embed_dims=[64, 128, 256, 512],
                 num_heads=[1, 2, 4, 8], mlp_ratios=[4, 4, 4, 4], qkv_bias=False, qk_scale=None, drop_rate=0.,
                 attn_drop_rate=0., drop_path_rate=0., norm_layer=nn.LayerNorm,
                 depths=[3, 4, 6, 3], sr_ratios=[8, 4, 2, 1]):
        super().__init__()
        self.num_classes = num_classes
        self.depths = depths

        # patch_embed
        self.patch_embed1 = OverlapPatchEmbed(img_size=img_size, patch_size=7, stride=4, in_channels=in_channels,
                                              embed_dim=embed_dims[0])
        self.patch_embed2 = OverlapPatchEmbed(img_size=img_size // 4, patch_size=3, stride=2, in_channels=embed_dims[0],
                                              embed_dim=embed_dims[1])
        self.patch_embed3 = OverlapPatchEmbed(img_size=img_size // 8, patch_size=3, stride=2, in_channels=embed_dims[1],
                                              embed_dim=embed_dims[2])
        self.patch_embed4 = OverlapPatchEmbed(img_size=img_size // 16, patch_size=3, stride=2, in_channels=embed_dims[2],
                                              embed_dim=embed_dims[3])

        # transformer encoder
        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))]  # stochastic depth decay rule
        cur = 0
        self.block1 = nn.ModuleList([Block(
            dim=embed_dims[0], num_heads=num_heads[0], mlp_ratio=mlp_ratios[0], qkv_bias=qkv_bias, qk_scale=qk_scale,
            drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[cur + i], norm_layer=norm_layer,
            sr_ratio=sr_ratios[0])
            for i in range(depths[0])])
        self.norm1 = norm_layer(embed_dims[0])

        cur += depths[0]
        self.block2 = nn.ModuleList([Block(
            dim=embed_dims[1], num_heads=num_heads[1], mlp_ratio=mlp_ratios[1], qkv_bias=qkv_bias, qk_scale=qk_scale,
            drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[cur + i], norm_layer=norm_layer,
            sr_ratio=sr_ratios[1])
            for i in range(depths[1])])
        self.norm2 = norm_layer(embed_dims[1])

        cur += depths[1]
        self.block3 = nn.ModuleList([Block(
            dim=embed_dims[2], num_heads=num_heads[2], mlp_ratio=mlp_ratios[2], qkv_bias=qkv_bias, qk_scale=qk_scale,
            drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[cur + i], norm_layer=norm_layer,
            sr_ratio=sr_ratios[2])
            for i in range(depths[2])])
        self.norm3 = norm_layer(embed_dims[2])

        cur += depths[2]
        self.block4 = nn.ModuleList([Block(
            dim=embed_dims[3], num_heads=num_heads[3], mlp_ratio=mlp_ratios[3], qkv_bias=qkv_bias, qk_scale=qk_scale,
            drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[cur + i], norm_layer=norm_layer,
            sr_ratio=sr_ratios[3])
            for i in range(depths[3])])
        self.norm4 = norm_layer(embed_dims[3])

        # classification head
        # self.head = nn.Linear(embed_dims[3], num_classes) if num_classes > 0 else nn.Identity()

        self.apply(self._init_weights)

    def _init_weights(self, m):
        if isinstance(m, nn.Linear):
            trunc_normal_(m.weight, std=.02)
            if isinstance(m, nn.Linear) and m.bias is not None:
                nn.init.constant_(m.bias, 0)
        elif isinstance(m, nn.LayerNorm):
            nn.init.constant_(m.bias, 0)
            nn.init.constant_(m.weight, 1.0)
        elif isinstance(m, nn.Conv2d):
            fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
            fan_out //= m.groups
            m.weight.data.normal_(0, math.sqrt(2.0 / fan_out))
            if m.bias is not None:
                m.bias.data.zero_()

    def init_weights(self, pretrained=None):
        if isinstance(pretrained, str):
            logger = 1
            #load_checkpoint(self, pretrained, map_location='cpu', strict=False, logger=logger)

    def reset_drop_path(self, drop_path_rate):
        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(self.depths))]
        cur = 0
        for i in range(self.depths[0]):
            self.block1[i].drop_path.drop_prob = dpr[cur + i]

        cur += self.depths[0]
        for i in range(self.depths[1]):
            self.block2[i].drop_path.drop_prob = dpr[cur + i]

        cur += self.depths[1]
        for i in range(self.depths[2]):
            self.block3[i].drop_path.drop_prob = dpr[cur + i]

        cur += self.depths[2]
        for i in range(self.depths[3]):
            self.block4[i].drop_path.drop_prob = dpr[cur + i]

    def freeze_patch_emb(self):
        self.patch_embed1.requires_grad = False

    @torch.jit.ignore
    def no_weight_decay(self):
        return {'pos_embed1', 'pos_embed2', 'pos_embed3', 'pos_embed4', 'cls_token'}  # has pos_embed may be better

    def get_classifier(self):
        return self.head

    def reset_classifier(self, num_classes, global_pool=''):
        self.num_classes = num_classes
        self.head = nn.Linear(self.embed_dim, num_classes) if num_classes > 0 else nn.Identity()

    def forward_features(self, x):
        B = x.shape[0]
        outs = []

        # stage 1
        x, H, W = self.patch_embed1(x)
        for i, blk in enumerate(self.block1):
            x = blk(x, H, W)
        x = self.norm1(x)
        x = x.reshape(B, H, W, -1).permute(0, 3, 1, 2).contiguous()
        outs.append(x)

        # stage 2
        x, H, W = self.patch_embed2(x)
        for i, blk in enumerate(self.block2):
            x = blk(x, H, W)
        x = self.norm2(x)
        x = x.reshape(B, H, W, -1).permute(0, 3, 1, 2).contiguous()
        outs.append(x)

        # stage 3
        x, H, W = self.patch_embed3(x)
        for i, blk in enumerate(self.block3):
            x = blk(x, H, W)
        x = self.norm3(x)
        x = x.reshape(B, H, W, -1).permute(0, 3, 1, 2).contiguous()
        outs.append(x)

        # stage 4
        x, H, W = self.patch_embed4(x)
        for i, blk in enumerate(self.block4):
            x = blk(x, H, W)
        x = self.norm4(x)
        x = x.reshape(B, H, W, -1).permute(0, 3, 1, 2).contiguous()
        outs.append(x)

        return outs

        # return x.mean(dim=1)

    def forward(self, x):
        x = self.forward_features(x)
        # x = self.head(x)

        return x


class DWConv(nn.Module):
    def __init__(self, dim=768):
        super(DWConv, self).__init__()
        self.dwconv = nn.Conv2d(dim, dim, 3, 1, 1, bias=True, groups=dim)

    def forward(self, x, H, W):
        B, N, C = x.shape
        x = x.transpose(1, 2).view(B, C, H, W).contiguous()
        x = self.dwconv(x)
        x = x.flatten(2).transpose(1, 2)

        return x


def _conv_filter(state_dict, patch_size=16):
    """ convert patch embedding weight from manual patchify + linear proj to conv"""
    out_dict = {}
    for k, v in state_dict.items():
        if 'patch_embed.proj.weight' in k:
            v = v.reshape((v.shape[0], 3, patch_size, patch_size))
        out_dict[k] = v

    return out_dict


## @register_model
class pvt_v2_b0(PyramidVisionTransformerImpr):
    def __init__(self, **kwargs):
        super(pvt_v2_b0, self).__init__(
            patch_size=4, embed_dims=[32, 64, 160, 256], num_heads=[1, 2, 5, 8], mlp_ratios=[8, 8, 4, 4],
            qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), depths=[2, 2, 2, 2], sr_ratios=[8, 4, 2, 1],
            drop_rate=0.0, drop_path_rate=0.1)



## @register_model
class pvt_v2_b1(PyramidVisionTransformerImpr):
    def __init__(self, **kwargs):
        super(pvt_v2_b1, self).__init__(
            patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[8, 8, 4, 4],
            qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), depths=[2, 2, 2, 2], sr_ratios=[8, 4, 2, 1],
            drop_rate=0.0, drop_path_rate=0.1)

## @register_model
class pvt_v2_b2(PyramidVisionTransformerImpr):
    def __init__(self, in_channels=3, **kwargs):
        super(pvt_v2_b2, self).__init__(
            patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[8, 8, 4, 4],
            qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), depths=[3, 4, 6, 3], sr_ratios=[8, 4, 2, 1],
            drop_rate=0.0, drop_path_rate=0.1, in_channels=in_channels)

## @register_model
class pvt_v2_b3(PyramidVisionTransformerImpr):
    def __init__(self, **kwargs):
        super(pvt_v2_b3, self).__init__(
            patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[8, 8, 4, 4],
            qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), depths=[3, 4, 18, 3], sr_ratios=[8, 4, 2, 1],
            drop_rate=0.0, drop_path_rate=0.1)

## @register_model
class pvt_v2_b4(PyramidVisionTransformerImpr):
    def __init__(self, **kwargs):
        super(pvt_v2_b4, self).__init__(
            patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[8, 8, 4, 4],
            qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), depths=[3, 8, 27, 3], sr_ratios=[8, 4, 2, 1],
            drop_rate=0.0, drop_path_rate=0.1)


## @register_model
class pvt_v2_b5(PyramidVisionTransformerImpr):
    def __init__(self, **kwargs):
        super(pvt_v2_b5, self).__init__(
            patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[4, 4, 4, 4],
            qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), depths=[3, 6, 40, 3], sr_ratios=[8, 4, 2, 1],
            drop_rate=0.0, drop_path_rate=0.1)



### models/backbones/swin_v1.py

# --------------------------------------------------------
# Swin Transformer
# Copyright (c) 2021 Microsoft
# Licensed under The MIT License [see LICENSE for details]
# Written by Ze Liu, Yutong Lin, Yixuan Wei
# --------------------------------------------------------

import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.checkpoint as checkpoint
import numpy as np
from timm.models.layers import DropPath, to_2tuple, trunc_normal_

# from config import Config


# config = Config()


class Mlp(nn.Module):
    """ Multilayer perceptron."""

    def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.):
        super().__init__()
        out_features = out_features or in_features
        hidden_features = hidden_features or in_features
        self.fc1 = nn.Linear(in_features, hidden_features)
        self.act = act_layer()
        self.fc2 = nn.Linear(hidden_features, out_features)
        self.drop = nn.Dropout(drop)

    def forward(self, x):
        x = self.fc1(x)
        x = self.act(x)
        x = self.drop(x)
        x = self.fc2(x)
        x = self.drop(x)
        return x


def window_partition(x, window_size):
    """
    Args:
        x: (B, H, W, C)
        window_size (int): window size

    Returns:
        windows: (num_windows*B, window_size, window_size, C)
    """
    B, H, W, C = x.shape
    x = x.view(B, H // window_size, window_size, W // window_size, window_size, C)
    windows = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(-1, window_size, window_size, C)
    return windows


def window_reverse(windows, window_size, H, W):
    """
    Args:
        windows: (num_windows*B, window_size, window_size, C)
        window_size (int): Window size
        H (int): Height of image
        W (int): Width of image

    Returns:
        x: (B, H, W, C)
    """
    B = int(windows.shape[0] / (H * W / window_size / window_size))
    x = windows.view(B, H // window_size, W // window_size, window_size, window_size, -1)
    x = x.permute(0, 1, 3, 2, 4, 5).contiguous().view(B, H, W, -1)
    return x


class WindowAttention(nn.Module):
    """ Window based multi-head self attention (W-MSA) module with relative position bias.
    It supports both of shifted and non-shifted window.

    Args:
        dim (int): Number of input channels.
        window_size (tuple[int]): The height and width of the window.
        num_heads (int): Number of attention heads.
        qkv_bias (bool, optional):  If True, add a learnable bias to query, key, value. Default: True
        qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set
        attn_drop (float, optional): Dropout ratio of attention weight. Default: 0.0
        proj_drop (float, optional): Dropout ratio of output. Default: 0.0
    """

    def __init__(self, dim, window_size, num_heads, qkv_bias=True, qk_scale=None, attn_drop=0., proj_drop=0.):

        super().__init__()
        self.dim = dim
        self.window_size = window_size  # Wh, Ww
        self.num_heads = num_heads
        head_dim = dim // num_heads
        self.scale = qk_scale or head_dim ** -0.5

        # define a parameter table of relative position bias
        self.relative_position_bias_table = nn.Parameter(
            torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1), num_heads))  # 2*Wh-1 * 2*Ww-1, nH

        # get pair-wise relative position index for each token inside the window
        coords_h = torch.arange(self.window_size[0])
        coords_w = torch.arange(self.window_size[1])
        coords = torch.stack(torch.meshgrid([coords_h, coords_w], indexing='ij'))  # 2, Wh, Ww
        coords_flatten = torch.flatten(coords, 1)  # 2, Wh*Ww
        relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :]  # 2, Wh*Ww, Wh*Ww
        relative_coords = relative_coords.permute(1, 2, 0).contiguous()  # Wh*Ww, Wh*Ww, 2
        relative_coords[:, :, 0] += self.window_size[0] - 1  # shift to start from 0
        relative_coords[:, :, 1] += self.window_size[1] - 1
        relative_coords[:, :, 0] *= 2 * self.window_size[1] - 1
        relative_position_index = relative_coords.sum(-1)  # Wh*Ww, Wh*Ww
        self.register_buffer("relative_position_index", relative_position_index)

        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
        self.attn_drop_prob = attn_drop
        self.attn_drop = nn.Dropout(attn_drop)
        self.proj = nn.Linear(dim, dim)
        self.proj_drop = nn.Dropout(proj_drop)

        trunc_normal_(self.relative_position_bias_table, std=.02)
        self.softmax = nn.Softmax(dim=-1)

    def forward(self, x, mask=None):
        """ Forward function.

        Args:
            x: input features with shape of (num_windows*B, N, C)
            mask: (0/-inf) mask with shape of (num_windows, Wh*Ww, Wh*Ww) or None
        """
        B_, N, C = x.shape
        qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
        q, k, v = qkv[0], qkv[1], qkv[2]  # make torchscript happy (cannot use tensor as tuple)

        q = q * self.scale

        if config.SDPA_enabled:
            x = torch.nn.functional.scaled_dot_product_attention(
                q, k, v,
                attn_mask=None, dropout_p=self.attn_drop_prob, is_causal=False
            ).transpose(1, 2).reshape(B_, N, C)
        else:
            attn = (q @ k.transpose(-2, -1))

            relative_position_bias = self.relative_position_bias_table[self.relative_position_index.view(-1)].view(
                self.window_size[0] * self.window_size[1], self.window_size[0] * self.window_size[1], -1
            )   # Wh*Ww, Wh*Ww, nH
            relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous()  # nH, Wh*Ww, Wh*Ww
            attn = attn + relative_position_bias.unsqueeze(0)

            if mask is not None:
                nW = mask.shape[0]
                attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0)
                attn = attn.view(-1, self.num_heads, N, N)
                attn = self.softmax(attn)
            else:
                attn = self.softmax(attn)

            attn = self.attn_drop(attn)

            x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        x = self.proj(x)
        x = self.proj_drop(x)
        return x


class SwinTransformerBlock(nn.Module):
    """ Swin Transformer Block.

    Args:
        dim (int): Number of input channels.
        num_heads (int): Number of attention heads.
        window_size (int): Window size.
        shift_size (int): Shift size for SW-MSA.
        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim.
        qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True
        qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set.
        drop (float, optional): Dropout rate. Default: 0.0
        attn_drop (float, optional): Attention dropout rate. Default: 0.0
        drop_path (float, optional): Stochastic depth rate. Default: 0.0
        act_layer (nn.Module, optional): Activation layer. Default: nn.GELU
        norm_layer (nn.Module, optional): Normalization layer.  Default: nn.LayerNorm
    """

    def __init__(self, dim, num_heads, window_size=7, shift_size=0,
                 mlp_ratio=4., qkv_bias=True, qk_scale=None, drop=0., attn_drop=0., drop_path=0.,
                 act_layer=nn.GELU, norm_layer=nn.LayerNorm):
        super().__init__()
        self.dim = dim
        self.num_heads = num_heads
        self.window_size = window_size
        self.shift_size = shift_size
        self.mlp_ratio = mlp_ratio
        assert 0 <= self.shift_size < self.window_size, "shift_size must in 0-window_size"

        self.norm1 = norm_layer(dim)
        self.attn = WindowAttention(
            dim, window_size=to_2tuple(self.window_size), num_heads=num_heads,
            qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)

        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
        self.norm2 = norm_layer(dim)
        mlp_hidden_dim = int(dim * mlp_ratio)
        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)

        self.H = None
        self.W = None

    def forward(self, x, mask_matrix):
        """ Forward function.

        Args:
            x: Input feature, tensor size (B, H*W, C).
            H, W: Spatial resolution of the input feature.
            mask_matrix: Attention mask for cyclic shift.
        """
        B, L, C = x.shape
        H, W = self.H, self.W
        assert L == H * W, "input feature has wrong size"

        shortcut = x
        x = self.norm1(x)
        x = x.view(B, H, W, C)

        # pad feature maps to multiples of window size
        pad_l = pad_t = 0
        pad_r = (self.window_size - W % self.window_size) % self.window_size
        pad_b = (self.window_size - H % self.window_size) % self.window_size
        x = F.pad(x, (0, 0, pad_l, pad_r, pad_t, pad_b))
        _, Hp, Wp, _ = x.shape

        # cyclic shift
        if self.shift_size > 0:
            shifted_x = torch.roll(x, shifts=(-self.shift_size, -self.shift_size), dims=(1, 2))
            attn_mask = mask_matrix
        else:
            shifted_x = x
            attn_mask = None

        # partition windows
        x_windows = window_partition(shifted_x, self.window_size)  # nW*B, window_size, window_size, C
        x_windows = x_windows.view(-1, self.window_size * self.window_size, C)  # nW*B, window_size*window_size, C

        # W-MSA/SW-MSA
        attn_windows = self.attn(x_windows, mask=attn_mask)  # nW*B, window_size*window_size, C

        # merge windows
        attn_windows = attn_windows.view(-1, self.window_size, self.window_size, C)
        shifted_x = window_reverse(attn_windows, self.window_size, Hp, Wp)  # B H' W' C

        # reverse cyclic shift
        if self.shift_size > 0:
            x = torch.roll(shifted_x, shifts=(self.shift_size, self.shift_size), dims=(1, 2))
        else:
            x = shifted_x

        if pad_r > 0 or pad_b > 0:
            x = x[:, :H, :W, :].contiguous()

        x = x.view(B, H * W, C)

        # FFN
        x = shortcut + self.drop_path(x)
        x = x + self.drop_path(self.mlp(self.norm2(x)))

        return x


class PatchMerging(nn.Module):
    """ Patch Merging Layer

    Args:
        dim (int): Number of input channels.
        norm_layer (nn.Module, optional): Normalization layer.  Default: nn.LayerNorm
    """
    def __init__(self, dim, norm_layer=nn.LayerNorm):
        super().__init__()
        self.dim = dim
        self.reduction = nn.Linear(4 * dim, 2 * dim, bias=False)
        self.norm = norm_layer(4 * dim)

    def forward(self, x, H, W):
        """ Forward function.

        Args:
            x: Input feature, tensor size (B, H*W, C).
            H, W: Spatial resolution of the input feature.
        """
        B, L, C = x.shape
        assert L == H * W, "input feature has wrong size"

        x = x.view(B, H, W, C)

        # padding
        pad_input = (H % 2 == 1) or (W % 2 == 1)
        if pad_input:
            x = F.pad(x, (0, 0, 0, W % 2, 0, H % 2))

        x0 = x[:, 0::2, 0::2, :]  # B H/2 W/2 C
        x1 = x[:, 1::2, 0::2, :]  # B H/2 W/2 C
        x2 = x[:, 0::2, 1::2, :]  # B H/2 W/2 C
        x3 = x[:, 1::2, 1::2, :]  # B H/2 W/2 C
        x = torch.cat([x0, x1, x2, x3], -1)  # B H/2 W/2 4*C
        x = x.view(B, -1, 4 * C)  # B H/2*W/2 4*C

        x = self.norm(x)
        x = self.reduction(x)

        return x


class BasicLayer(nn.Module):
    """ A basic Swin Transformer layer for one stage.

    Args:
        dim (int): Number of feature channels
        depth (int): Depths of this stage.
        num_heads (int): Number of attention head.
        window_size (int): Local window size. Default: 7.
        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4.
        qkv_bias (bool, optional): If True, add a learnable bias to query, key, value. Default: True
        qk_scale (float | None, optional): Override default qk scale of head_dim ** -0.5 if set.
        drop (float, optional): Dropout rate. Default: 0.0
        attn_drop (float, optional): Attention dropout rate. Default: 0.0
        drop_path (float | tuple[float], optional): Stochastic depth rate. Default: 0.0
        norm_layer (nn.Module, optional): Normalization layer. Default: nn.LayerNorm
        downsample (nn.Module | None, optional): Downsample layer at the end of the layer. Default: None
        use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False.
    """

    def __init__(self,
                 dim,
                 depth,
                 num_heads,
                 window_size=7,
                 mlp_ratio=4.,
                 qkv_bias=True,
                 qk_scale=None,
                 drop=0.,
                 attn_drop=0.,
                 drop_path=0.,
                 norm_layer=nn.LayerNorm,
                 downsample=None,
                 use_checkpoint=False):
        super().__init__()
        self.window_size = window_size
        self.shift_size = window_size // 2
        self.depth = depth
        self.use_checkpoint = use_checkpoint

        # build blocks
        self.blocks = nn.ModuleList([
            SwinTransformerBlock(
                dim=dim,
                num_heads=num_heads,
                window_size=window_size,
                shift_size=0 if (i % 2 == 0) else window_size // 2,
                mlp_ratio=mlp_ratio,
                qkv_bias=qkv_bias,
                qk_scale=qk_scale,
                drop=drop,
                attn_drop=attn_drop,
                drop_path=drop_path[i] if isinstance(drop_path, list) else drop_path,
                norm_layer=norm_layer)
            for i in range(depth)])

        # patch merging layer
        if downsample is not None:
            self.downsample = downsample(dim=dim, norm_layer=norm_layer)
        else:
            self.downsample = None

    def forward(self, x, H, W):
        """ Forward function.

        Args:
            x: Input feature, tensor size (B, H*W, C).
            H, W: Spatial resolution of the input feature.
        """

        # calculate attention mask for SW-MSA
        # Turn int to torch.tensor for the compatiability with torch.compile in PyTorch 2.5.
        Hp = torch.ceil(torch.tensor(H) / self.window_size).to(torch.int64) * self.window_size
        Wp = torch.ceil(torch.tensor(W) / self.window_size).to(torch.int64) * self.window_size
        img_mask = torch.zeros((1, Hp, Wp, 1), device=x.device)  # 1 Hp Wp 1
        h_slices = (slice(0, -self.window_size),
                    slice(-self.window_size, -self.shift_size),
                    slice(-self.shift_size, None))
        w_slices = (slice(0, -self.window_size),
                    slice(-self.window_size, -self.shift_size),
                    slice(-self.shift_size, None))
        cnt = 0
        for h in h_slices:
            for w in w_slices:
                img_mask[:, h, w, :] = cnt
                cnt += 1

        mask_windows = window_partition(img_mask, self.window_size)  # nW, window_size, window_size, 1
        mask_windows = mask_windows.view(-1, self.window_size * self.window_size)
        attn_mask = mask_windows.unsqueeze(1) - mask_windows.unsqueeze(2)
        attn_mask = attn_mask.masked_fill(attn_mask != 0, float(-100.0)).masked_fill(attn_mask == 0, float(0.0)).to(x.dtype)

        for blk in self.blocks:
            blk.H, blk.W = H, W
            if self.use_checkpoint:
                x = checkpoint.checkpoint(blk, x, attn_mask)
            else:
                x = blk(x, attn_mask)
        if self.downsample is not None:
            x_down = self.downsample(x, H, W)
            Wh, Ww = (H + 1) // 2, (W + 1) // 2
            return x, H, W, x_down, Wh, Ww
        else:
            return x, H, W, x, H, W


class PatchEmbed(nn.Module):
    """ Image to Patch Embedding

    Args:
        patch_size (int): Patch token size. Default: 4.
        in_channels (int): Number of input image channels. Default: 3.
        embed_dim (int): Number of linear projection output channels. Default: 96.
        norm_layer (nn.Module, optional): Normalization layer. Default: None
    """

    def __init__(self, patch_size=4, in_channels=3, embed_dim=96, norm_layer=None):
        super().__init__()
        patch_size = to_2tuple(patch_size)
        self.patch_size = patch_size

        self.in_channels = in_channels
        self.embed_dim = embed_dim

        self.proj = nn.Conv2d(in_channels, embed_dim, kernel_size=patch_size, stride=patch_size)
        if norm_layer is not None:
            self.norm = norm_layer(embed_dim)
        else:
            self.norm = None

    def forward(self, x):
        """Forward function."""
        # padding
        _, _, H, W = x.size()
        if W % self.patch_size[1] != 0:
            x = F.pad(x, (0, self.patch_size[1] - W % self.patch_size[1]))
        if H % self.patch_size[0] != 0:
            x = F.pad(x, (0, 0, 0, self.patch_size[0] - H % self.patch_size[0]))

        x = self.proj(x)  # B C Wh Ww
        if self.norm is not None:
            Wh, Ww = x.size(2), x.size(3)
            x = x.flatten(2).transpose(1, 2)
            x = self.norm(x)
            x = x.transpose(1, 2).view(-1, self.embed_dim, Wh, Ww)

        return x


class SwinTransformer(nn.Module):
    """ Swin Transformer backbone.
        A PyTorch impl of : `Swin Transformer: Hierarchical Vision Transformer using Shifted Windows`  -
          https://arxiv.org/pdf/2103.14030

    Args:
        pretrain_img_size (int): Input image size for training the pretrained model,
            used in absolute postion embedding. Default 224.
        patch_size (int | tuple(int)): Patch size. Default: 4.
        in_channels (int): Number of input image channels. Default: 3.
        embed_dim (int): Number of linear projection output channels. Default: 96.
        depths (tuple[int]): Depths of each Swin Transformer stage.
        num_heads (tuple[int]): Number of attention head of each stage.
        window_size (int): Window size. Default: 7.
        mlp_ratio (float): Ratio of mlp hidden dim to embedding dim. Default: 4.
        qkv_bias (bool): If True, add a learnable bias to query, key, value. Default: True
        qk_scale (float): Override default qk scale of head_dim ** -0.5 if set.
        drop_rate (float): Dropout rate.
        attn_drop_rate (float): Attention dropout rate. Default: 0.
        drop_path_rate (float): Stochastic depth rate. Default: 0.2.
        norm_layer (nn.Module): Normalization layer. Default: nn.LayerNorm.
        ape (bool): If True, add absolute position embedding to the patch embedding. Default: False.
        patch_norm (bool): If True, add normalization after patch embedding. Default: True.
        out_indices (Sequence[int]): Output from which stages.
        frozen_stages (int): Stages to be frozen (stop grad and set eval mode).
            -1 means not freezing any parameters.
        use_checkpoint (bool): Whether to use checkpointing to save memory. Default: False.
    """

    def __init__(self,
                 pretrain_img_size=224,
                 patch_size=4,
                 in_channels=3,
                 embed_dim=96,
                 depths=[2, 2, 6, 2],
                 num_heads=[3, 6, 12, 24],
                 window_size=7,
                 mlp_ratio=4.,
                 qkv_bias=True,
                 qk_scale=None,
                 drop_rate=0.,
                 attn_drop_rate=0.,
                 drop_path_rate=0.2,
                 norm_layer=nn.LayerNorm,
                 ape=False,
                 patch_norm=True,
                 out_indices=(0, 1, 2, 3),
                 frozen_stages=-1,
                 use_checkpoint=False):
        super().__init__()

        self.pretrain_img_size = pretrain_img_size
        self.num_layers = len(depths)
        self.embed_dim = embed_dim
        self.ape = ape
        self.patch_norm = patch_norm
        self.out_indices = out_indices
        self.frozen_stages = frozen_stages

        # split image into non-overlapping patches
        self.patch_embed = PatchEmbed(
            patch_size=patch_size, in_channels=in_channels, embed_dim=embed_dim,
            norm_layer=norm_layer if self.patch_norm else None)

        # absolute position embedding
        if self.ape:
            pretrain_img_size = to_2tuple(pretrain_img_size)
            patch_size = to_2tuple(patch_size)
            patches_resolution = [pretrain_img_size[0] // patch_size[0], pretrain_img_size[1] // patch_size[1]]

            self.absolute_pos_embed = nn.Parameter(torch.zeros(1, embed_dim, patches_resolution[0], patches_resolution[1]))
            trunc_normal_(self.absolute_pos_embed, std=.02)

        self.pos_drop = nn.Dropout(p=drop_rate)

        # stochastic depth
        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))]  # stochastic depth decay rule

        # build layers
        self.layers = nn.ModuleList()
        for i_layer in range(self.num_layers):
            layer = BasicLayer(
                dim=int(embed_dim * 2 ** i_layer),
                depth=depths[i_layer],
                num_heads=num_heads[i_layer],
                window_size=window_size,
                mlp_ratio=mlp_ratio,
                qkv_bias=qkv_bias,
                qk_scale=qk_scale,
                drop=drop_rate,
                attn_drop=attn_drop_rate,
                drop_path=dpr[sum(depths[:i_layer]):sum(depths[:i_layer + 1])],
                norm_layer=norm_layer,
                downsample=PatchMerging if (i_layer < self.num_layers - 1) else None,
                use_checkpoint=use_checkpoint)
            self.layers.append(layer)

        num_features = [int(embed_dim * 2 ** i) for i in range(self.num_layers)]
        self.num_features = num_features

        # add a norm layer for each output
        for i_layer in out_indices:
            layer = norm_layer(num_features[i_layer])
            layer_name = f'norm{i_layer}'
            self.add_module(layer_name, layer)

        self._freeze_stages()

    def _freeze_stages(self):
        if self.frozen_stages >= 0:
            self.patch_embed.eval()
            for param in self.patch_embed.parameters():
                param.requires_grad = False

        if self.frozen_stages >= 1 and self.ape:
            self.absolute_pos_embed.requires_grad = False

        if self.frozen_stages >= 2:
            self.pos_drop.eval()
            for i in range(0, self.frozen_stages - 1):
                m = self.layers[i]
                m.eval()
                for param in m.parameters():
                    param.requires_grad = False


    def forward(self, x):
        """Forward function."""
        x = self.patch_embed(x)

        Wh, Ww = x.size(2), x.size(3)
        if self.ape:
            # interpolate the position embedding to the corresponding size
            absolute_pos_embed = F.interpolate(self.absolute_pos_embed, size=(Wh, Ww), mode='bicubic')
            x = (x + absolute_pos_embed) # B Wh*Ww C
            
        outs = []#x.contiguous()]
        x = x.flatten(2).transpose(1, 2)
        x = self.pos_drop(x)
        for i in range(self.num_layers):
            layer = self.layers[i]
            x_out, H, W, x, Wh, Ww = layer(x, Wh, Ww)

            if i in self.out_indices:
                norm_layer = getattr(self, f'norm{i}')
                x_out = norm_layer(x_out)

                out = x_out.view(-1, H, W, self.num_features[i]).permute(0, 3, 1, 2).contiguous()
                outs.append(out)

        return tuple(outs)

    def train(self, mode=True):
        """Convert the model into training mode while keep layers freezed."""
        super(SwinTransformer, self).train(mode)
        self._freeze_stages()

def swin_v1_t():
    model = SwinTransformer(embed_dim=96, depths=[2, 2, 6, 2], num_heads=[3, 6, 12, 24], window_size=7)
    return model

def swin_v1_s():
    model = SwinTransformer(embed_dim=96, depths=[2, 2, 18, 2], num_heads=[3, 6, 12, 24], window_size=7)
    return model

def swin_v1_b():
    model = SwinTransformer(embed_dim=128, depths=[2, 2, 18, 2], num_heads=[4, 8, 16, 32], window_size=12)
    return model

def swin_v1_l():
    model = SwinTransformer(embed_dim=192, depths=[2, 2, 18, 2], num_heads=[6, 12, 24, 48], window_size=12)
    return model



### models/modules/deform_conv.py

import torch
import torch.nn as nn
from torchvision.ops import deform_conv2d


class DeformableConv2d(nn.Module):
    def __init__(self,
                 in_channels,
                 out_channels,
                 kernel_size=3,
                 stride=1,
                 padding=1,
                 bias=False):

        super(DeformableConv2d, self).__init__()
        
        assert type(kernel_size) == tuple or type(kernel_size) == int

        kernel_size = kernel_size if type(kernel_size) == tuple else (kernel_size, kernel_size)
        self.stride = stride if type(stride) == tuple else (stride, stride)
        self.padding = padding
        
        self.offset_conv = nn.Conv2d(in_channels,
                                     2 * kernel_size[0] * kernel_size[1],
                                     kernel_size=kernel_size,
                                     stride=stride,
                                     padding=self.padding,
                                     bias=True)

        nn.init.constant_(self.offset_conv.weight, 0.)
        nn.init.constant_(self.offset_conv.bias, 0.)
        
        self.modulator_conv = nn.Conv2d(in_channels,
                                     1 * kernel_size[0] * kernel_size[1],
                                     kernel_size=kernel_size,
                                     stride=stride,
                                     padding=self.padding,
                                     bias=True)

        nn.init.constant_(self.modulator_conv.weight, 0.)
        nn.init.constant_(self.modulator_conv.bias, 0.)

        self.regular_conv = nn.Conv2d(in_channels,
                                      out_channels=out_channels,
                                      kernel_size=kernel_size,
                                      stride=stride,
                                      padding=self.padding,
                                      bias=bias)

    def forward(self, x):
        #h, w = x.shape[2:]
        #max_offset = max(h, w)/4.

        offset = self.offset_conv(x)#.clamp(-max_offset, max_offset)
        modulator = 2. * torch.sigmoid(self.modulator_conv(x))
        
        x = deform_conv2d(
            input=x,
            offset=offset,
            weight=self.regular_conv.weight,
            bias=self.regular_conv.bias,
            padding=self.padding,
            mask=modulator,
            stride=self.stride,
        )
        return x




### utils.py

import torch.nn as nn


def build_act_layer(act_layer):
    if act_layer == 'ReLU':
        return nn.ReLU(inplace=True)
    elif act_layer == 'SiLU':
        return nn.SiLU(inplace=True)
    elif act_layer == 'GELU':
        return nn.GELU()

    raise NotImplementedError(f'build_act_layer does not support {act_layer}')


def build_norm_layer(dim,
                     norm_layer,
                     in_format='channels_last',
                     out_format='channels_last',
                     eps=1e-6):
    layers = []
    if norm_layer == 'BN':
        if in_format == 'channels_last':
            layers.append(to_channels_first())
        layers.append(nn.BatchNorm2d(dim))
        if out_format == 'channels_last':
            layers.append(to_channels_last())
    elif norm_layer == 'LN':
        if in_format == 'channels_first':
            layers.append(to_channels_last())
        layers.append(nn.LayerNorm(dim, eps=eps))
        if out_format == 'channels_first':
            layers.append(to_channels_first())
    else:
        raise NotImplementedError(
            f'build_norm_layer does not support {norm_layer}')
    return nn.Sequential(*layers)


class to_channels_first(nn.Module):

    def __init__(self):
        super().__init__()

    def forward(self, x):
        return x.permute(0, 3, 1, 2)


class to_channels_last(nn.Module):

    def __init__(self):
        super().__init__()

    def forward(self, x):
        return x.permute(0, 2, 3, 1)



### dataset.py

_class_labels_TR_sorted = (
    'Airplane, Ant, Antenna, Archery, Axe, BabyCarriage, Bag, BalanceBeam, Balcony, Balloon, Basket, BasketballHoop, Beatle, Bed, Bee, Bench, Bicycle, '
    'BicycleFrame, BicycleStand, Boat, Bonsai, BoomLift, Bridge, BunkBed, Butterfly, Button, Cable, CableLift, Cage, Camcorder, Cannon, Canoe, Car, '
    'CarParkDropArm, Carriage, Cart, Caterpillar, CeilingLamp, Centipede, Chair, Clip, Clock, Clothes, CoatHanger, Comb, ConcretePumpTruck, Crack, Crane, '
    'Cup, DentalChair, Desk, DeskChair, Diagram, DishRack, DoorHandle, Dragonfish, Dragonfly, Drum, Earphone, Easel, ElectricIron, Excavator, Eyeglasses, '
    'Fan, Fence, Fencing, FerrisWheel, FireExtinguisher, Fishing, Flag, FloorLamp, Forklift, GasStation, Gate, Gear, Goal, Golf, GymEquipment, Hammock, '
    'Handcart, Handcraft, Handrail, HangGlider, Harp, Harvester, Headset, Helicopter, Helmet, Hook, HorizontalBar, Hydrovalve, IroningTable, Jewelry, Key, '
    'KidsPlayground, Kitchenware, Kite, Knife, Ladder, LaundryRack, Lightning, Lobster, Locust, Machine, MachineGun, MagazineRack, Mantis, Medal, MemorialArchway, '
    'Microphone, Missile, MobileHolder, Monitor, Mosquito, Motorcycle, MovingTrolley, Mower, MusicPlayer, MusicStand, ObservationTower, Octopus, OilWell, '
    'OlympicLogo, OperatingTable, OutdoorFitnessEquipment, Parachute, Pavilion, Piano, Pipe, PlowHarrow, PoleVault, Punchbag, Rack, Racket, Rifle, Ring, Robot, '
    'RockClimbing, Rope, Sailboat, Satellite, Scaffold, Scale, Scissor, Scooter, Sculpture, Seadragon, Seahorse, Seal, SewingMachine, Ship, Shoe, ShoppingCart, '
    'ShoppingTrolley, Shower, Shrimp, Signboard, Skateboarding, Skeleton, Skiing, Spade, SpeedBoat, Spider, Spoon, Stair, Stand, Stationary, SteeringWheel, '
    'Stethoscope, Stool, Stove, StreetLamp, SweetStand, Swing, Sword, TV, Table, TableChair, TableLamp, TableTennis, Tank, Tapeline, Teapot, Telescope, Tent, '
    'TobaccoPipe, Toy, Tractor, TrafficLight, TrafficSign, Trampoline, TransmissionTower, Tree, Tricycle, TrimmerCover, Tripod, Trombone, Truck, Trumpet, Tuba, '
    'UAV, Umbrella, UnevenBars, UtilityPole, VacuumCleaner, Violin, Wakesurfing, Watch, WaterTower, WateringPot, Well, WellLid, Wheel, Wheelchair, WindTurbine, Windmill, WineGlass, WireWhisk, Yacht'
)
class_labels_TR_sorted = _class_labels_TR_sorted.split(', ')


### models/backbones/build_backbones.py

import torch
import torch.nn as nn
from collections import OrderedDict
from torchvision.models import vgg16, vgg16_bn, VGG16_Weights, VGG16_BN_Weights, resnet50, ResNet50_Weights
# from models.pvt_v2 import pvt_v2_b0, pvt_v2_b1, pvt_v2_b2, pvt_v2_b5
# from models.swin_v1 import swin_v1_t, swin_v1_s, swin_v1_b, swin_v1_l
# from config import Config


config = Config()

def build_backbone(bb_name, pretrained=True, params_settings=''):
    if bb_name == 'vgg16':
        bb_net = list(vgg16(pretrained=VGG16_Weights.DEFAULT if pretrained else None).children())[0]
        bb = nn.Sequential(OrderedDict({'conv1': bb_net[:4], 'conv2': bb_net[4:9], 'conv3': bb_net[9:16], 'conv4': bb_net[16:23]}))
    elif bb_name == 'vgg16bn':
        bb_net = list(vgg16_bn(pretrained=VGG16_BN_Weights.DEFAULT if pretrained else None).children())[0]
        bb = nn.Sequential(OrderedDict({'conv1': bb_net[:6], 'conv2': bb_net[6:13], 'conv3': bb_net[13:23], 'conv4': bb_net[23:33]}))
    elif bb_name == 'resnet50':
        bb_net = list(resnet50(pretrained=ResNet50_Weights.DEFAULT if pretrained else None).children())
        bb = nn.Sequential(OrderedDict({'conv1': nn.Sequential(*bb_net[0:3]), 'conv2': bb_net[4], 'conv3': bb_net[5], 'conv4': bb_net[6]}))
    else:
        bb = eval('{}({})'.format(bb_name, params_settings))
        if pretrained:
            bb = load_weights(bb, bb_name)
    return bb

def load_weights(model, model_name):
    save_model = torch.load(config.weights[model_name], map_location='cpu')
    model_dict = model.state_dict()
    state_dict = {k: v if v.size() == model_dict[k].size() else model_dict[k] for k, v in save_model.items() if k in model_dict.keys()}
    # to ignore the weights with mismatched size when I modify the backbone itself.
    if not state_dict:
        save_model_keys = list(save_model.keys())
        sub_item = save_model_keys[0] if len(save_model_keys) == 1 else None
        state_dict = {k: v if v.size() == model_dict[k].size() else model_dict[k] for k, v in save_model[sub_item].items() if k in model_dict.keys()}
        if not state_dict or not sub_item:
            print('Weights are not successully loaded. Check the state dict of weights file.')
            return None
        else:
            print('Found correct weights in the "{}" item of loaded state_dict.'.format(sub_item))
    model_dict.update(state_dict)
    model.load_state_dict(model_dict)
    return model



### models/modules/decoder_blocks.py

import torch
import torch.nn as nn
# from models.aspp import ASPP, ASPPDeformable
# from config import Config


# config = Config()


class BasicDecBlk(nn.Module):
    def __init__(self, in_channels=64, out_channels=64, inter_channels=64):
        super(BasicDecBlk, self).__init__()
        inter_channels = in_channels // 4 if config.dec_channels_inter == 'adap' else 64
        self.conv_in = nn.Conv2d(in_channels, inter_channels, 3, 1, padding=1)
        self.relu_in = nn.ReLU(inplace=True)
        if config.dec_att == 'ASPP':
            self.dec_att = ASPP(in_channels=inter_channels)
        elif config.dec_att == 'ASPPDeformable':
            self.dec_att = ASPPDeformable(in_channels=inter_channels)
        self.conv_out = nn.Conv2d(inter_channels, out_channels, 3, 1, padding=1)
        self.bn_in = nn.BatchNorm2d(inter_channels) if config.batch_size > 1 else nn.Identity()
        self.bn_out = nn.BatchNorm2d(out_channels) if config.batch_size > 1 else nn.Identity()

    def forward(self, x):
        x = self.conv_in(x)
        x = self.bn_in(x)
        x = self.relu_in(x)
        if hasattr(self, 'dec_att'):
            x = self.dec_att(x)
        x = self.conv_out(x)
        x = self.bn_out(x)
        return x


class ResBlk(nn.Module):
    def __init__(self, in_channels=64, out_channels=None, inter_channels=64):
        super(ResBlk, self).__init__()
        if out_channels is None:
            out_channels = in_channels
        inter_channels = in_channels // 4 if config.dec_channels_inter == 'adap' else 64

        self.conv_in = nn.Conv2d(in_channels, inter_channels, 3, 1, padding=1)
        self.bn_in = nn.BatchNorm2d(inter_channels) if config.batch_size > 1 else nn.Identity()
        self.relu_in = nn.ReLU(inplace=True)

        if config.dec_att == 'ASPP':
            self.dec_att = ASPP(in_channels=inter_channels)
        elif config.dec_att == 'ASPPDeformable':
            self.dec_att = ASPPDeformable(in_channels=inter_channels)

        self.conv_out = nn.Conv2d(inter_channels, out_channels, 3, 1, padding=1)
        self.bn_out = nn.BatchNorm2d(out_channels) if config.batch_size > 1 else nn.Identity()
        
        self.conv_resi = nn.Conv2d(in_channels, out_channels, 1, 1, 0)

    def forward(self, x):
        _x = self.conv_resi(x)
        x = self.conv_in(x)
        x = self.bn_in(x)
        x = self.relu_in(x)
        if hasattr(self, 'dec_att'):
            x = self.dec_att(x)
        x = self.conv_out(x)
        x = self.bn_out(x)
        return x + _x



### models/modules/lateral_blocks.py

import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
from functools import partial

# from config import Config


# config = Config()


class BasicLatBlk(nn.Module):
    def __init__(self, in_channels=64, out_channels=64, inter_channels=64):
        super(BasicLatBlk, self).__init__()
        inter_channels = in_channels // 4 if config.dec_channels_inter == 'adap' else 64
        self.conv = nn.Conv2d(in_channels, out_channels, 1, 1, 0)

    def forward(self, x):
        x = self.conv(x)
        return x



### models/modules/aspp.py

import torch
import torch.nn as nn
import torch.nn.functional as F
# from models.deform_conv import DeformableConv2d
# from config import Config


# config = Config()


class _ASPPModule(nn.Module):
    def __init__(self, in_channels, planes, kernel_size, padding, dilation):
        super(_ASPPModule, self).__init__()
        self.atrous_conv = nn.Conv2d(in_channels, planes, kernel_size=kernel_size,
                                            stride=1, padding=padding, dilation=dilation, bias=False)
        self.bn = nn.BatchNorm2d(planes) if config.batch_size > 1 else nn.Identity()
        self.relu = nn.ReLU(inplace=True)

    def forward(self, x):
        x = self.atrous_conv(x)
        x = self.bn(x)

        return self.relu(x)


class ASPP(nn.Module):
    def __init__(self, in_channels=64, out_channels=None, output_stride=16):
        super(ASPP, self).__init__()
        self.down_scale = 1
        if out_channels is None:
            out_channels = in_channels
        self.in_channelster = 256 // self.down_scale
        if output_stride == 16:
            dilations = [1, 6, 12, 18]
        elif output_stride == 8:
            dilations = [1, 12, 24, 36]
        else:
            raise NotImplementedError

        self.aspp1 = _ASPPModule(in_channels, self.in_channelster, 1, padding=0, dilation=dilations[0])
        self.aspp2 = _ASPPModule(in_channels, self.in_channelster, 3, padding=dilations[1], dilation=dilations[1])
        self.aspp3 = _ASPPModule(in_channels, self.in_channelster, 3, padding=dilations[2], dilation=dilations[2])
        self.aspp4 = _ASPPModule(in_channels, self.in_channelster, 3, padding=dilations[3], dilation=dilations[3])

        self.global_avg_pool = nn.Sequential(nn.AdaptiveAvgPool2d((1, 1)),
                                             nn.Conv2d(in_channels, self.in_channelster, 1, stride=1, bias=False),
                                             nn.BatchNorm2d(self.in_channelster) if config.batch_size > 1 else nn.Identity(),
                                             nn.ReLU(inplace=True))
        self.conv1 = nn.Conv2d(self.in_channelster * 5, out_channels, 1, bias=False)
        self.bn1 = nn.BatchNorm2d(out_channels) if config.batch_size > 1 else nn.Identity()
        self.relu = nn.ReLU(inplace=True)
        self.dropout = nn.Dropout(0.5)

    def forward(self, x):
        x1 = self.aspp1(x)
        x2 = self.aspp2(x)
        x3 = self.aspp3(x)
        x4 = self.aspp4(x)
        x5 = self.global_avg_pool(x)
        x5 = F.interpolate(x5, size=x1.size()[2:], mode='bilinear', align_corners=True)
        x = torch.cat((x1, x2, x3, x4, x5), dim=1)

        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)

        return self.dropout(x)


##################### Deformable
class _ASPPModuleDeformable(nn.Module):
    def __init__(self, in_channels, planes, kernel_size, padding):
        super(_ASPPModuleDeformable, self).__init__()
        self.atrous_conv = DeformableConv2d(in_channels, planes, kernel_size=kernel_size,
                                            stride=1, padding=padding, bias=False)
        self.bn = nn.BatchNorm2d(planes) if config.batch_size > 1 else nn.Identity()
        self.relu = nn.ReLU(inplace=True)

    def forward(self, x):
        x = self.atrous_conv(x)
        x = self.bn(x)

        return self.relu(x)


class ASPPDeformable(nn.Module):
    def __init__(self, in_channels, out_channels=None, parallel_block_sizes=[1, 3, 7]):
        super(ASPPDeformable, self).__init__()
        self.down_scale = 1
        if out_channels is None:
            out_channels = in_channels
        self.in_channelster = 256 // self.down_scale

        self.aspp1 = _ASPPModuleDeformable(in_channels, self.in_channelster, 1, padding=0)
        self.aspp_deforms = nn.ModuleList([
            _ASPPModuleDeformable(in_channels, self.in_channelster, conv_size, padding=int(conv_size//2)) for conv_size in parallel_block_sizes
        ])

        self.global_avg_pool = nn.Sequential(nn.AdaptiveAvgPool2d((1, 1)),
                                             nn.Conv2d(in_channels, self.in_channelster, 1, stride=1, bias=False),
                                             nn.BatchNorm2d(self.in_channelster) if config.batch_size > 1 else nn.Identity(),
                                             nn.ReLU(inplace=True))
        self.conv1 = nn.Conv2d(self.in_channelster * (2 + len(self.aspp_deforms)), out_channels, 1, bias=False)
        self.bn1 = nn.BatchNorm2d(out_channels) if config.batch_size > 1 else nn.Identity()
        self.relu = nn.ReLU(inplace=True)
        self.dropout = nn.Dropout(0.5)

    def forward(self, x):
        x1 = self.aspp1(x)
        x_aspp_deforms = [aspp_deform(x) for aspp_deform in self.aspp_deforms]
        x5 = self.global_avg_pool(x)
        x5 = F.interpolate(x5, size=x1.size()[2:], mode='bilinear', align_corners=True)
        x = torch.cat((x1, *x_aspp_deforms, x5), dim=1)

        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)

        return self.dropout(x)



### models/refinement/refiner.py

import torch
import torch.nn as nn
from collections import OrderedDict
import torch
import torch.nn as nn
import torch.nn.functional as F
from torchvision.models import vgg16, vgg16_bn
from torchvision.models import resnet50

# from config import Config
# from dataset import class_labels_TR_sorted
# from models.build_backbone import build_backbone
# from models.decoder_blocks import BasicDecBlk
# from models.lateral_blocks import BasicLatBlk
# from models.ing import *
# from models.stem_layer import StemLayer


class RefinerPVTInChannels4(nn.Module):
    def __init__(self, in_channels=3+1):
        super(RefinerPVTInChannels4, self).__init__()
        self.config = Config()
        self.epoch = 1
        self.bb = build_backbone(self.config.bb, params_settings='in_channels=4')

        lateral_channels_in_collection = {
            'vgg16': [512, 256, 128, 64], 'vgg16bn': [512, 256, 128, 64], 'resnet50': [1024, 512, 256, 64],
            'pvt_v2_b2': [512, 320, 128, 64], 'pvt_v2_b5': [512, 320, 128, 64],
            'swin_v1_b': [1024, 512, 256, 128], 'swin_v1_l': [1536, 768, 384, 192],
        }
        channels = lateral_channels_in_collection[self.config.bb]
        self.squeeze_module = BasicDecBlk(channels[0], channels[0])

        self.decoder = Decoder(channels)

        if 0:
            for key, value in self.named_parameters():
                if 'bb.' in key:
                    value.requires_grad = False

    def forward(self, x):
        if isinstance(x, list):
            x = torch.cat(x, dim=1)
        ########## Encoder ##########
        if self.config.bb in ['vgg16', 'vgg16bn', 'resnet50']:
            x1 = self.bb.conv1(x)
            x2 = self.bb.conv2(x1)
            x3 = self.bb.conv3(x2)
            x4 = self.bb.conv4(x3)
        else:
            x1, x2, x3, x4 = self.bb(x)

        x4 = self.squeeze_module(x4)

        ########## Decoder ##########

        features = [x, x1, x2, x3, x4]
        scaled_preds = self.decoder(features)

        return scaled_preds


class Refiner(nn.Module):
    def __init__(self, in_channels=3+1):
        super(Refiner, self).__init__()
        self.config = Config()
        self.epoch = 1
        self.stem_layer = StemLayer(in_channels=in_channels, inter_channels=48, out_channels=3, norm_layer='BN' if self.config.batch_size > 1 else 'LN')
        self.bb = build_backbone(self.config.bb)

        lateral_channels_in_collection = {
            'vgg16': [512, 256, 128, 64], 'vgg16bn': [512, 256, 128, 64], 'resnet50': [1024, 512, 256, 64],
            'pvt_v2_b2': [512, 320, 128, 64], 'pvt_v2_b5': [512, 320, 128, 64],
            'swin_v1_b': [1024, 512, 256, 128], 'swin_v1_l': [1536, 768, 384, 192],
        }
        channels = lateral_channels_in_collection[self.config.bb]
        self.squeeze_module = BasicDecBlk(channels[0], channels[0])

        self.decoder = Decoder(channels)

        if 0:
            for key, value in self.named_parameters():
                if 'bb.' in key:
                    value.requires_grad = False

    def forward(self, x):
        if isinstance(x, list):
            x = torch.cat(x, dim=1)
        x = self.stem_layer(x)
        ########## Encoder ##########
        if self.config.bb in ['vgg16', 'vgg16bn', 'resnet50']:
            x1 = self.bb.conv1(x)
            x2 = self.bb.conv2(x1)
            x3 = self.bb.conv3(x2)
            x4 = self.bb.conv4(x3)
        else:
            x1, x2, x3, x4 = self.bb(x)

        x4 = self.squeeze_module(x4)

        ########## Decoder ##########

        features = [x, x1, x2, x3, x4]
        scaled_preds = self.decoder(features)

        return scaled_preds


class Decoder(nn.Module):
    def __init__(self, channels):
        super(Decoder, self).__init__()
        self.config = Config()
        DecoderBlock = eval('BasicDecBlk')
        LateralBlock = eval('BasicLatBlk')

        self.decoder_block4 = DecoderBlock(channels[0], channels[1])
        self.decoder_block3 = DecoderBlock(channels[1], channels[2])
        self.decoder_block2 = DecoderBlock(channels[2], channels[3])
        self.decoder_block1 = DecoderBlock(channels[3], channels[3]//2)

        self.lateral_block4 = LateralBlock(channels[1], channels[1])
        self.lateral_block3 = LateralBlock(channels[2], channels[2])
        self.lateral_block2 = LateralBlock(channels[3], channels[3])

        if self.config.ms_supervision:
            self.conv_ms_spvn_4 = nn.Conv2d(channels[1], 1, 1, 1, 0)
            self.conv_ms_spvn_3 = nn.Conv2d(channels[2], 1, 1, 1, 0)
            self.conv_ms_spvn_2 = nn.Conv2d(channels[3], 1, 1, 1, 0)
        self.conv_out1 = nn.Sequential(nn.Conv2d(channels[3]//2, 1, 1, 1, 0))

    def forward(self, features):
        x, x1, x2, x3, x4 = features
        outs = []
        p4 = self.decoder_block4(x4)
        _p4 = F.interpolate(p4, size=x3.shape[2:], mode='bilinear', align_corners=True)
        _p3 = _p4 + self.lateral_block4(x3)

        p3 = self.decoder_block3(_p3)
        _p3 = F.interpolate(p3, size=x2.shape[2:], mode='bilinear', align_corners=True)
        _p2 = _p3 + self.lateral_block3(x2)

        p2 = self.decoder_block2(_p2)
        _p2 = F.interpolate(p2, size=x1.shape[2:], mode='bilinear', align_corners=True)
        _p1 = _p2 + self.lateral_block2(x1)

        _p1 = self.decoder_block1(_p1)
        _p1 = F.interpolate(_p1, size=x.shape[2:], mode='bilinear', align_corners=True)
        p1_out = self.conv_out1(_p1)

        if self.config.ms_supervision:
            outs.append(self.conv_ms_spvn_4(p4))
            outs.append(self.conv_ms_spvn_3(p3))
            outs.append(self.conv_ms_spvn_2(p2))
        outs.append(p1_out)
        return outs


class RefUNet(nn.Module):
    # Refinement
    def __init__(self, in_channels=3+1):
        super(RefUNet, self).__init__()
        self.encoder_1 = nn.Sequential(
            nn.Conv2d(in_channels, 64, 3, 1, 1),
            nn.Conv2d(64, 64, 3, 1, 1),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True)
        )

        self.encoder_2 = nn.Sequential(
            nn.MaxPool2d(2, 2, ceil_mode=True),
            nn.Conv2d(64, 64, 3, 1, 1),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True)
        )

        self.encoder_3 = nn.Sequential(
            nn.MaxPool2d(2, 2, ceil_mode=True),
            nn.Conv2d(64, 64, 3, 1, 1),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True)
        )

        self.encoder_4 = nn.Sequential(
            nn.MaxPool2d(2, 2, ceil_mode=True),
            nn.Conv2d(64, 64, 3, 1, 1),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True)
        )

        self.pool4 = nn.MaxPool2d(2, 2, ceil_mode=True)
        #####
        self.decoder_5 = nn.Sequential(
            nn.Conv2d(64, 64, 3, 1, 1),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True)
        )
        #####
        self.decoder_4 = nn.Sequential(
            nn.Conv2d(128, 64, 3, 1, 1),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True)
        )

        self.decoder_3 = nn.Sequential(
            nn.Conv2d(128, 64, 3, 1, 1),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True)
        )

        self.decoder_2 = nn.Sequential(
            nn.Conv2d(128, 64, 3, 1, 1),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True)
        )

        self.decoder_1 = nn.Sequential(
            nn.Conv2d(128, 64, 3, 1, 1),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True)
        )

        self.conv_d0 = nn.Conv2d(64, 1, 3, 1, 1)

        self.upscore2 = nn.Upsample(scale_factor=2, mode='bilinear', align_corners=True)

    def forward(self, x):
        outs = []
        if isinstance(x, list):
            x = torch.cat(x, dim=1)
        hx = x

        hx1 = self.encoder_1(hx)
        hx2 = self.encoder_2(hx1)
        hx3 = self.encoder_3(hx2)
        hx4 = self.encoder_4(hx3)

        hx = self.decoder_5(self.pool4(hx4))
        hx = torch.cat((self.upscore2(hx), hx4), 1)

        d4 = self.decoder_4(hx)
        hx = torch.cat((self.upscore2(d4), hx3), 1)

        d3 = self.decoder_3(hx)
        hx = torch.cat((self.upscore2(d3), hx2), 1)

        d2 = self.decoder_2(hx)
        hx = torch.cat((self.upscore2(d2), hx1), 1)

        d1 = self.decoder_1(hx)

        x = self.conv_d0(d1)
        outs.append(x)
        return outs



### models/stem_layer.py

import torch.nn as nn
# from utils import build_act_layer, build_norm_layer


class StemLayer(nn.Module):
    r""" Stem layer of InternImage
    Args:
        in_channels (int): number of input channels
        out_channels (int): number of output channels
        act_layer (str): activation layer
        norm_layer (str): normalization layer
    """

    def __init__(self,
                 in_channels=3+1,
                 inter_channels=48,
                 out_channels=96,
                 act_layer='GELU',
                 norm_layer='BN'):
        super().__init__()
        self.conv1 = nn.Conv2d(in_channels,
                               inter_channels,
                               kernel_size=3,
                               stride=1,
                               padding=1)
        self.norm1 = build_norm_layer(
            inter_channels, norm_layer, 'channels_first', 'channels_first'
        )
        self.act = build_act_layer(act_layer)
        self.conv2 = nn.Conv2d(inter_channels,
                               out_channels,
                               kernel_size=3,
                               stride=1,
                               padding=1)
        self.norm2 = build_norm_layer(
            out_channels, norm_layer, 'channels_first', 'channels_first'
        )

    def forward(self, x):
        x = self.conv1(x)
        x = self.norm1(x)
        x = self.act(x)
        x = self.conv2(x)
        x = self.norm2(x)
        return x


### models/birefnet.py

import torch
import torch.nn as nn
import torch.nn.functional as F
from kornia.filters import laplacian
from transformers import PreTrainedModel
from einops import rearrange

# from config import Config
# from dataset import class_labels_TR_sorted
# from models.build_backbone import build_backbone
# from models.decoder_blocks import BasicDecBlk, ResBlk, HierarAttDecBlk
# from models.lateral_blocks import BasicLatBlk
# from models.aspp import ASPP, ASPPDeformable
# from models.ing import *
# from models.refiner import Refiner, RefinerPVTInChannels4, RefUNet
# from models.stem_layer import StemLayer
from .BiRefNet_config import BiRefNetConfig


def image2patches(image, grid_h=2, grid_w=2, patch_ref=None, transformation='b c (hg h) (wg w) -> (b hg wg) c h w'):
    if patch_ref is not None:
        grid_h, grid_w = image.shape[-2] // patch_ref.shape[-2], image.shape[-1] // patch_ref.shape[-1]
    patches = rearrange(image, transformation, hg=grid_h, wg=grid_w)
    return patches

def patches2image(patches, grid_h=2, grid_w=2, patch_ref=None, transformation='(b hg wg) c h w -> b c (hg h) (wg w)'):
    if patch_ref is not None:
        grid_h, grid_w = patch_ref.shape[-2] // patches[0].shape[-2], patch_ref.shape[-1] // patches[0].shape[-1]
    image = rearrange(patches, transformation, hg=grid_h, wg=grid_w)
    return image

class BiRefNet(
    PreTrainedModel
):
    config_class = BiRefNetConfig
    def __init__(self, bb_pretrained=True, config=BiRefNetConfig()):
        super(BiRefNet, self).__init__(config)
        bb_pretrained = config.bb_pretrained
        self.config = Config()
        self.epoch = 1
        self.bb = build_backbone(self.config.bb, pretrained=bb_pretrained)

        channels = self.config.lateral_channels_in_collection

        if self.config.auxiliary_classification:
            self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
            self.cls_head = nn.Sequential(
                nn.Linear(channels[0], len(class_labels_TR_sorted))
            )

        if self.config.squeeze_block:
            self.squeeze_module = nn.Sequential(*[
                eval(self.config.squeeze_block.split('_x')[0])(channels[0]+sum(self.config.cxt), channels[0])
                for _ in range(eval(self.config.squeeze_block.split('_x')[1]))
            ])

        self.decoder = Decoder(channels)

        if self.config.ender:
            self.dec_end = nn.Sequential(
                nn.Conv2d(1, 16, 3, 1, 1),
                nn.Conv2d(16, 1, 3, 1, 1),
                nn.ReLU(inplace=True),
            )

        # refine patch-level segmentation
        if self.config.refine:
            if self.config.refine == 'itself':
                self.stem_layer = StemLayer(in_channels=3+1, inter_channels=48, out_channels=3, norm_layer='BN' if self.config.batch_size > 1 else 'LN')
            else:
                self.refiner = eval('{}({})'.format(self.config.refine, 'in_channels=3+1'))

        if self.config.freeze_bb:
            # Freeze the backbone...
            print(self.named_parameters())
            for key, value in self.named_parameters():
                if 'bb.' in key and 'refiner.' not in key:
                    value.requires_grad = False

    def forward_enc(self, x):
        if self.config.bb in ['vgg16', 'vgg16bn', 'resnet50']:
            x1 = self.bb.conv1(x); x2 = self.bb.conv2(x1); x3 = self.bb.conv3(x2); x4 = self.bb.conv4(x3)
        else:
            x1, x2, x3, x4 = self.bb(x)
            if self.config.mul_scl_ipt == 'cat':
                B, C, H, W = x.shape
                x1_, x2_, x3_, x4_ = self.bb(F.interpolate(x, size=(H//2, W//2), mode='bilinear', align_corners=True))
                x1 = torch.cat([x1, F.interpolate(x1_, size=x1.shape[2:], mode='bilinear', align_corners=True)], dim=1)
                x2 = torch.cat([x2, F.interpolate(x2_, size=x2.shape[2:], mode='bilinear', align_corners=True)], dim=1)
                x3 = torch.cat([x3, F.interpolate(x3_, size=x3.shape[2:], mode='bilinear', align_corners=True)], dim=1)
                x4 = torch.cat([x4, F.interpolate(x4_, size=x4.shape[2:], mode='bilinear', align_corners=True)], dim=1)
            elif self.config.mul_scl_ipt == 'add':
                B, C, H, W = x.shape
                x1_, x2_, x3_, x4_ = self.bb(F.interpolate(x, size=(H//2, W//2), mode='bilinear', align_corners=True))
                x1 = x1 + F.interpolate(x1_, size=x1.shape[2:], mode='bilinear', align_corners=True)
                x2 = x2 + F.interpolate(x2_, size=x2.shape[2:], mode='bilinear', align_corners=True)
                x3 = x3 + F.interpolate(x3_, size=x3.shape[2:], mode='bilinear', align_corners=True)
                x4 = x4 + F.interpolate(x4_, size=x4.shape[2:], mode='bilinear', align_corners=True)
        class_preds = self.cls_head(self.avgpool(x4).view(x4.shape[0], -1)) if self.training and self.config.auxiliary_classification else None
        if self.config.cxt:
            x4 = torch.cat(
                (
                    *[
                        F.interpolate(x1, size=x4.shape[2:], mode='bilinear', align_corners=True),
                        F.interpolate(x2, size=x4.shape[2:], mode='bilinear', align_corners=True),
                        F.interpolate(x3, size=x4.shape[2:], mode='bilinear', align_corners=True),
                    ][-len(self.config.cxt):],
                    x4
                ),
                dim=1
            )
        return (x1, x2, x3, x4), class_preds

    def forward_ori(self, x):
        ########## Encoder ##########
        (x1, x2, x3, x4), class_preds = self.forward_enc(x)
        if self.config.squeeze_block:
            x4 = self.squeeze_module(x4)
        ########## Decoder ##########
        features = [x, x1, x2, x3, x4]
        if self.training and self.config.out_ref:
            features.append(laplacian(torch.mean(x, dim=1).unsqueeze(1), kernel_size=5))
        scaled_preds = self.decoder(features)
        return scaled_preds, class_preds

    def forward(self, x):
        scaled_preds, class_preds = self.forward_ori(x)
        class_preds_lst = [class_preds]
        return [scaled_preds, class_preds_lst] if self.training else scaled_preds


class Decoder(nn.Module):
    def __init__(self, channels):
        super(Decoder, self).__init__()
        self.config = Config()
        DecoderBlock = eval(self.config.dec_blk)
        LateralBlock = eval(self.config.lat_blk)

        if self.config.dec_ipt:
            self.split = self.config.dec_ipt_split
            N_dec_ipt = 64
            DBlock = SimpleConvs
            ic = 64
            ipt_cha_opt = 1
            self.ipt_blk5 = DBlock(2**10*3 if self.split else 3, [N_dec_ipt, channels[0]//8][ipt_cha_opt], inter_channels=ic)
            self.ipt_blk4 = DBlock(2**8*3 if self.split else 3, [N_dec_ipt, channels[0]//8][ipt_cha_opt], inter_channels=ic)
            self.ipt_blk3 = DBlock(2**6*3 if self.split else 3, [N_dec_ipt, channels[1]//8][ipt_cha_opt], inter_channels=ic)
            self.ipt_blk2 = DBlock(2**4*3 if self.split else 3, [N_dec_ipt, channels[2]//8][ipt_cha_opt], inter_channels=ic)
            self.ipt_blk1 = DBlock(2**0*3 if self.split else 3, [N_dec_ipt, channels[3]//8][ipt_cha_opt], inter_channels=ic)
        else:
            self.split = None

        self.decoder_block4 = DecoderBlock(channels[0]+([N_dec_ipt, channels[0]//8][ipt_cha_opt] if self.config.dec_ipt else 0), channels[1])
        self.decoder_block3 = DecoderBlock(channels[1]+([N_dec_ipt, channels[0]//8][ipt_cha_opt] if self.config.dec_ipt else 0), channels[2])
        self.decoder_block2 = DecoderBlock(channels[2]+([N_dec_ipt, channels[1]//8][ipt_cha_opt] if self.config.dec_ipt else 0), channels[3])
        self.decoder_block1 = DecoderBlock(channels[3]+([N_dec_ipt, channels[2]//8][ipt_cha_opt] if self.config.dec_ipt else 0), channels[3]//2)
        self.conv_out1 = nn.Sequential(nn.Conv2d(channels[3]//2+([N_dec_ipt, channels[3]//8][ipt_cha_opt] if self.config.dec_ipt else 0), 1, 1, 1, 0))

        self.lateral_block4 = LateralBlock(channels[1], channels[1])
        self.lateral_block3 = LateralBlock(channels[2], channels[2])
        self.lateral_block2 = LateralBlock(channels[3], channels[3])

        if self.config.ms_supervision:
            self.conv_ms_spvn_4 = nn.Conv2d(channels[1], 1, 1, 1, 0)
            self.conv_ms_spvn_3 = nn.Conv2d(channels[2], 1, 1, 1, 0)
            self.conv_ms_spvn_2 = nn.Conv2d(channels[3], 1, 1, 1, 0)

            if self.config.out_ref:
                _N = 16
                self.gdt_convs_4 = nn.Sequential(nn.Conv2d(channels[1], _N, 3, 1, 1), nn.BatchNorm2d(_N) if self.config.batch_size > 1 else nn.Identity(), nn.ReLU(inplace=True))
                self.gdt_convs_3 = nn.Sequential(nn.Conv2d(channels[2], _N, 3, 1, 1), nn.BatchNorm2d(_N) if self.config.batch_size > 1 else nn.Identity(), nn.ReLU(inplace=True))
                self.gdt_convs_2 = nn.Sequential(nn.Conv2d(channels[3], _N, 3, 1, 1), nn.BatchNorm2d(_N) if self.config.batch_size > 1 else nn.Identity(), nn.ReLU(inplace=True))

                self.gdt_convs_pred_4 = nn.Sequential(nn.Conv2d(_N, 1, 1, 1, 0))
                self.gdt_convs_pred_3 = nn.Sequential(nn.Conv2d(_N, 1, 1, 1, 0))
                self.gdt_convs_pred_2 = nn.Sequential(nn.Conv2d(_N, 1, 1, 1, 0))
                
                self.gdt_convs_attn_4 = nn.Sequential(nn.Conv2d(_N, 1, 1, 1, 0))
                self.gdt_convs_attn_3 = nn.Sequential(nn.Conv2d(_N, 1, 1, 1, 0))
                self.gdt_convs_attn_2 = nn.Sequential(nn.Conv2d(_N, 1, 1, 1, 0))

    def forward(self, features):
        if self.training and self.config.out_ref:
            outs_gdt_pred = []
            outs_gdt_label = []
            x, x1, x2, x3, x4, gdt_gt = features
        else:
            x, x1, x2, x3, x4 = features
        outs = []

        if self.config.dec_ipt:
            patches_batch = image2patches(x, patch_ref=x4, transformation='b c (hg h) (wg w) -> b (c hg wg) h w') if self.split else x
            x4 = torch.cat((x4, self.ipt_blk5(F.interpolate(patches_batch, size=x4.shape[2:], mode='bilinear', align_corners=True))), 1)
        p4 = self.decoder_block4(x4)
        m4 = self.conv_ms_spvn_4(p4) if self.config.ms_supervision and self.training else None
        if self.config.out_ref:
            p4_gdt = self.gdt_convs_4(p4)
            if self.training:
                # >> GT:
                m4_dia = m4
                gdt_label_main_4 = gdt_gt * F.interpolate(m4_dia, size=gdt_gt.shape[2:], mode='bilinear', align_corners=True)
                outs_gdt_label.append(gdt_label_main_4)
                # >> Pred:
                gdt_pred_4 = self.gdt_convs_pred_4(p4_gdt)
                outs_gdt_pred.append(gdt_pred_4)
            gdt_attn_4 = self.gdt_convs_attn_4(p4_gdt).sigmoid()
            # >> Finally:
            p4 = p4 * gdt_attn_4
        _p4 = F.interpolate(p4, size=x3.shape[2:], mode='bilinear', align_corners=True)
        _p3 = _p4 + self.lateral_block4(x3)

        if self.config.dec_ipt:
            patches_batch = image2patches(x, patch_ref=_p3, transformation='b c (hg h) (wg w) -> b (c hg wg) h w') if self.split else x
            _p3 = torch.cat((_p3, self.ipt_blk4(F.interpolate(patches_batch, size=x3.shape[2:], mode='bilinear', align_corners=True))), 1)
        p3 = self.decoder_block3(_p3)
        m3 = self.conv_ms_spvn_3(p3) if self.config.ms_supervision and self.training else None
        if self.config.out_ref:
            p3_gdt = self.gdt_convs_3(p3)
            if self.training:
                # >> GT:
                # m3 --dilation--> m3_dia
                # G_3^gt * m3_dia --> G_3^m, which is the label of gradient
                m3_dia = m3
                gdt_label_main_3 = gdt_gt * F.interpolate(m3_dia, size=gdt_gt.shape[2:], mode='bilinear', align_corners=True)
                outs_gdt_label.append(gdt_label_main_3)
                # >> Pred:
                # p3 --conv--BN--> F_3^G, where F_3^G predicts the \hat{G_3} with xx
                # F_3^G --sigmoid--> A_3^G
                gdt_pred_3 = self.gdt_convs_pred_3(p3_gdt)
                outs_gdt_pred.append(gdt_pred_3)
            gdt_attn_3 = self.gdt_convs_attn_3(p3_gdt).sigmoid()
            # >> Finally:
            # p3 = p3 * A_3^G
            p3 = p3 * gdt_attn_3
        _p3 = F.interpolate(p3, size=x2.shape[2:], mode='bilinear', align_corners=True)
        _p2 = _p3 + self.lateral_block3(x2)

        if self.config.dec_ipt:
            patches_batch = image2patches(x, patch_ref=_p2, transformation='b c (hg h) (wg w) -> b (c hg wg) h w') if self.split else x
            _p2 = torch.cat((_p2, self.ipt_blk3(F.interpolate(patches_batch, size=x2.shape[2:], mode='bilinear', align_corners=True))), 1)
        p2 = self.decoder_block2(_p2)
        m2 = self.conv_ms_spvn_2(p2) if self.config.ms_supervision and self.training else None
        if self.config.out_ref:
            p2_gdt = self.gdt_convs_2(p2)
            if self.training:
                # >> GT:
                m2_dia = m2
                gdt_label_main_2 = gdt_gt * F.interpolate(m2_dia, size=gdt_gt.shape[2:], mode='bilinear', align_corners=True)
                outs_gdt_label.append(gdt_label_main_2)
                # >> Pred:
                gdt_pred_2 = self.gdt_convs_pred_2(p2_gdt)
                outs_gdt_pred.append(gdt_pred_2)
            gdt_attn_2 = self.gdt_convs_attn_2(p2_gdt).sigmoid()
            # >> Finally:
            p2 = p2 * gdt_attn_2
        _p2 = F.interpolate(p2, size=x1.shape[2:], mode='bilinear', align_corners=True)
        _p1 = _p2 + self.lateral_block2(x1)

        if self.config.dec_ipt:
            patches_batch = image2patches(x, patch_ref=_p1, transformation='b c (hg h) (wg w) -> b (c hg wg) h w') if self.split else x
            _p1 = torch.cat((_p1, self.ipt_blk2(F.interpolate(patches_batch, size=x1.shape[2:], mode='bilinear', align_corners=True))), 1)
        _p1 = self.decoder_block1(_p1)
        _p1 = F.interpolate(_p1, size=x.shape[2:], mode='bilinear', align_corners=True)

        if self.config.dec_ipt:
            patches_batch = image2patches(x, patch_ref=_p1, transformation='b c (hg h) (wg w) -> b (c hg wg) h w') if self.split else x
            _p1 = torch.cat((_p1, self.ipt_blk1(F.interpolate(patches_batch, size=x.shape[2:], mode='bilinear', align_corners=True))), 1)
        p1_out = self.conv_out1(_p1)

        if self.config.ms_supervision and self.training:
            outs.append(m4)
            outs.append(m3)
            outs.append(m2)
        outs.append(p1_out)
        return outs if not (self.config.out_ref and self.training) else ([outs_gdt_pred, outs_gdt_label], outs)


class SimpleConvs(nn.Module):
    def __init__(
        self, in_channels: int, out_channels: int, inter_channels=64
    ) -> None:
        super().__init__()
        self.conv1 = nn.Conv2d(in_channels, inter_channels, 3, 1, 1)
        self.conv_out = nn.Conv2d(inter_channels, out_channels, 3, 1, 1)

    def forward(self, x):
        return self.conv_out(self.conv1(x))