Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -3,8 +3,16 @@ from transformers import AutoModelForCausalLM, AutoTokenizer
|
|
3 |
from PIL import Image
|
4 |
import numpy as np
|
5 |
import os
|
|
|
6 |
import gradio as gr
|
7 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
# Load the model and tokenizer
|
9 |
model_path = "ByteDance/Sa2VA-4B"
|
10 |
|
@@ -20,6 +28,17 @@ tokenizer = AutoTokenizer.from_pretrained(
|
|
20 |
trust_remote_code = True,
|
21 |
)
|
22 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
23 |
def image_vision(image_input_path, prompt):
|
24 |
image_path = image_input_path
|
25 |
text_prompts = f"<image>{prompt}"
|
@@ -34,6 +53,7 @@ def image_vision(image_input_path, prompt):
|
|
34 |
return_dict = model.predict_forward(**input_dict)
|
35 |
print(return_dict)
|
36 |
answer = return_dict["prediction"] # the text format answer
|
|
|
37 |
seg_image = return_dict["prediction_masks"]
|
38 |
|
39 |
return answer, seg_image
|
@@ -41,7 +61,15 @@ def image_vision(image_input_path, prompt):
|
|
41 |
def main_infer(image_input_path, prompt):
|
42 |
|
43 |
answer, seg_image = image_vision(image_input_path, prompt)
|
44 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
45 |
|
46 |
# Gradio UI
|
47 |
|
@@ -56,7 +84,7 @@ with gr.Blocks() as demo:
|
|
56 |
submit_btn = gr.Button("Submit", scale=1)
|
57 |
with gr.Column():
|
58 |
output_res = gr.Textbox(label="Response")
|
59 |
-
output_image = gr.Image(label="Segmentation")
|
60 |
|
61 |
submit_btn.click(
|
62 |
fn = main_infer,
|
|
|
3 |
from PIL import Image
|
4 |
import numpy as np
|
5 |
import os
|
6 |
+
import tempfile
|
7 |
import gradio as gr
|
8 |
|
9 |
+
import cv2
|
10 |
+
try:
|
11 |
+
from mmengine.visualization import Visualizer
|
12 |
+
except ImportError:
|
13 |
+
Visualizer = None
|
14 |
+
print("Warning: mmengine is not installed, visualization is disabled.")
|
15 |
+
|
16 |
# Load the model and tokenizer
|
17 |
model_path = "ByteDance/Sa2VA-4B"
|
18 |
|
|
|
28 |
trust_remote_code = True,
|
29 |
)
|
30 |
|
31 |
+
def visualize(pred_mask, image_path, work_dir):
|
32 |
+
visualizer = Visualizer()
|
33 |
+
img = cv2.imread(image_path)
|
34 |
+
visualizer.set_image(img)
|
35 |
+
visualizer.draw_binary_masks(pred_mask, colors='g', alphas=0.4)
|
36 |
+
visual_result = visualizer.get_image()
|
37 |
+
|
38 |
+
output_path = os.path.join(work_dir, os.path.basename(image_path))
|
39 |
+
cv2.imwrite(output_path, visual_result)
|
40 |
+
return output_path
|
41 |
+
|
42 |
def image_vision(image_input_path, prompt):
|
43 |
image_path = image_input_path
|
44 |
text_prompts = f"<image>{prompt}"
|
|
|
53 |
return_dict = model.predict_forward(**input_dict)
|
54 |
print(return_dict)
|
55 |
answer = return_dict["prediction"] # the text format answer
|
56 |
+
|
57 |
seg_image = return_dict["prediction_masks"]
|
58 |
|
59 |
return answer, seg_image
|
|
|
61 |
def main_infer(image_input_path, prompt):
|
62 |
|
63 |
answer, seg_image = image_vision(image_input_path, prompt)
|
64 |
+
pred_masks = seg_image[0]
|
65 |
+
|
66 |
+
if '[SEG]' in answer and Visualizer is not None:
|
67 |
+
temp_dir = tempfile.mkdtemp()
|
68 |
+
pred_mask = pred_masks[0]
|
69 |
+
os.makedirs(temp_dir, exist_ok=True)
|
70 |
+
seg_result = visualize(pred_mask, image_input_path, temp_dir)
|
71 |
+
|
72 |
+
return answer, seg_result
|
73 |
|
74 |
# Gradio UI
|
75 |
|
|
|
84 |
submit_btn = gr.Button("Submit", scale=1)
|
85 |
with gr.Column():
|
86 |
output_res = gr.Textbox(label="Response")
|
87 |
+
output_image = gr.Image(label="Segmentation", type="numpy")
|
88 |
|
89 |
submit_btn.click(
|
90 |
fn = main_infer,
|