app.py

import streamlit as st
import torch
import torch.nn as nn
import timm
import torchvision.transforms as transforms
import pytorch_lightning as pl
from PIL import Image
import numpy as np
from torch import nn
import smp

# The accompanying inference app

PATHS = ['1.tiff', '2.tiff']


NUM_CLASSES = len(CLASSES)

IDS_TO_CLASSES_DICT = dict(zip(list(range(NUM_CLASSES)), CLASSES))


MODEL_NAME = "se_resne"
MODEL_PATH = "model.ckpt"
DEVICE = "cuda:0" if torch.cuda.is_available() else "cpu"
TRANSFORM = transforms.Compose([transforms.ToTensor(), 
                                transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])



BACKBONE = ""
IN_CHANNELS = ""
CLASSES = ""
# TODO: path to weights?
WEIGHTS = ""

class VesuviusModel(nn.Module):
    def __init__(self, weight=None):
        super().__init__()
        self.cfg = cfg

        self.encoder = smp.Unet(
            encoder_name=BACKBONE, 
            encoder_weights=WEIGHTS,
            in_channels=IN_CHANNELS,
            classes=CLASSES,
            activation=None,
        )

    def forward(self, image):
        output = self.encoder(image)
        output = output.squeeze(-1)
        return output


def load_weights_into_model(model_name: str, model_path: str) -> nn.Module:
    model = VesuviusModel(model_name)
    state_dict = torch.load(model_path, map_location=DEVICE)["state_dict"]
    model.load_state_dict(state_dict)
    return model








model = load_weights_into_model(MODEL_NAME, MODEL_PATH)
model.to(DEVICE)
model.eval()

img_path = st.selectbox('Select an image to segment', PATHS)

st.write('You have selected:', img_path)
img = Image.open(img_path)

st.image(img, caption='Selected image to segment')

np_img = np.array(img)

input_batch = TRANSFORM(np_img[:, :, :3]).unsqueeze(0).to(DEVICE)

with st.spinner("Segmenting the image in progress..."):


    with torch.no_grad():
        # TODO: Finish...
        prediction = model(input_batch).cpu()
        print(prediction)