Spaces:
Running
Running
update app.py
Browse files
app.py
CHANGED
@@ -6,66 +6,101 @@ Combined Medical-VLM, **SAM-2 automatic masking**, and CheXagent demo.
|
|
6 |
|
7 |
β Changes β
|
8 |
-----------
|
9 |
-
1.
|
10 |
-
2.
|
11 |
-
3.
|
12 |
-
4.
|
13 |
"""
|
14 |
|
15 |
# ---------------------------------------------------------------------
|
16 |
# Standard libs
|
17 |
# ---------------------------------------------------------------------
|
18 |
-
# ---------------------------------------------------------------------
|
19 |
-
import os, warnings
|
20 |
-
os.environ["PYTORCH_ENABLE_MPS_FALLBACK"] = "1" # CPU fallback for missing MPS ops
|
21 |
-
warnings.filterwarnings("ignore", message=r".*upsample_bicubic2d.*") # hide one-line notice
|
22 |
-
|
23 |
import os
|
24 |
import sys
|
25 |
import uuid
|
26 |
import tempfile
|
|
|
|
|
27 |
from threading import Thread
|
28 |
|
|
|
|
|
|
|
|
|
29 |
# ---------------------------------------------------------------------
|
30 |
# Third-party libs
|
31 |
-
|
32 |
-
|
33 |
# ---------------------------------------------------------------------
|
34 |
import torch
|
35 |
import numpy as np
|
36 |
from PIL import Image, ImageDraw
|
37 |
import gradio as gr
|
38 |
|
39 |
-
# If you cloned facebookresearch/sam2 into the repo root, make sure it's importable
|
40 |
-
sys.path.append(os.path.abspath("."))
|
41 |
-
|
42 |
# =============================================================================
|
43 |
-
#
|
44 |
# =============================================================================
|
45 |
-
|
46 |
-
|
47 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
48 |
|
49 |
# =============================================================================
|
50 |
-
# SAM-2 imports
|
51 |
# =============================================================================
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
|
|
|
|
|
|
|
|
|
|
|
62 |
|
63 |
# =============================================================================
|
64 |
# CheXagent imports
|
65 |
# =============================================================================
|
66 |
from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer
|
67 |
|
68 |
-
|
69 |
# ---------------------------------------------------------------------
|
70 |
# Devices
|
71 |
# ---------------------------------------------------------------------
|
@@ -76,7 +111,6 @@ def get_device():
|
|
76 |
return torch.device("mps")
|
77 |
return torch.device("cpu")
|
78 |
|
79 |
-
|
80 |
# =============================================================================
|
81 |
# Qwen-VLM model & agent
|
82 |
# =============================================================================
|
@@ -84,7 +118,6 @@ _qwen_model = None
|
|
84 |
_qwen_processor = None
|
85 |
_qwen_device = None
|
86 |
|
87 |
-
|
88 |
def load_qwen_model_and_processor(hf_token=None):
|
89 |
global _qwen_model, _qwen_processor, _qwen_device
|
90 |
if _qwen_model is None:
|
@@ -107,7 +140,6 @@ def load_qwen_model_and_processor(hf_token=None):
|
|
107 |
)
|
108 |
return _qwen_model, _qwen_processor, _qwen_device
|
109 |
|
110 |
-
|
111 |
class MedicalVLMAgent:
|
112 |
"""Light wrapper around Qwen-VLM with an optional image."""
|
113 |
|
@@ -150,56 +182,76 @@ class MedicalVLMAgent:
|
|
150 |
trimmed = out[0][inputs.input_ids.shape[1] :]
|
151 |
return self.processor.decode(trimmed, skip_special_tokens=True).strip()
|
152 |
|
153 |
-
|
154 |
-
# =============================================================================
|
155 |
-
# SAM-2 model + AutomaticMaskGenerator
|
156 |
-
# =============================================================================
|
157 |
-
|
158 |
-
# =============================================================================
|
159 |
-
# SAM-2.1 model + AutomaticMaskGenerator (concise version)
|
160 |
# =============================================================================
|
|
|
161 |
# =============================================================================
|
162 |
-
|
163 |
-
|
164 |
-
|
165 |
-
|
166 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
167 |
|
168 |
def initialize_sam2():
|
169 |
-
|
170 |
-
|
171 |
-
|
172 |
-
|
173 |
-
|
174 |
-
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
|
180 |
-
|
181 |
-
|
182 |
-
|
183 |
-
|
184 |
-
|
185 |
-
|
186 |
-
|
187 |
-
|
188 |
-
|
189 |
-
|
190 |
-
|
191 |
-
|
192 |
-
|
193 |
-
return sam2_model, mask_gen
|
194 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
195 |
|
196 |
-
#
|
197 |
-
|
|
|
198 |
_sam2_model, _mask_generator = initialize_sam2()
|
199 |
-
|
200 |
-
|
201 |
-
|
202 |
-
|
203 |
|
204 |
def automatic_mask_overlay(image_np: np.ndarray) -> np.ndarray:
|
205 |
"""Generate masks and alpha-blend them on top of the original image."""
|
@@ -222,9 +274,13 @@ def automatic_mask_overlay(image_np: np.ndarray) -> np.ndarray:
|
|
222 |
return overlay
|
223 |
|
224 |
def tumor_segmentation_interface(image: Image.Image | None):
|
|
|
225 |
if image is None:
|
226 |
return None, "Please upload an image."
|
227 |
|
|
|
|
|
|
|
228 |
if _mask_generator is None:
|
229 |
return None, "SAM-2 not properly initialized. Check the console for errors."
|
230 |
|
@@ -237,30 +293,81 @@ def tumor_segmentation_interface(image: Image.Image | None):
|
|
237 |
return None, f"SAM-2 error: {e}"
|
238 |
|
239 |
# =============================================================================
|
240 |
-
#
|
241 |
# =============================================================================
|
242 |
-
|
243 |
-
|
244 |
-
|
245 |
-
|
246 |
-
|
247 |
-
|
248 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
249 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
250 |
|
251 |
def get_model_device(model):
|
|
|
|
|
252 |
for p in model.parameters():
|
253 |
return p.device
|
254 |
return torch.device("cpu")
|
255 |
|
256 |
-
|
257 |
def clean_text(text):
|
258 |
return text.replace("</s>", "")
|
259 |
|
260 |
-
|
261 |
@torch.no_grad()
|
262 |
def response_report_generation(pil_image_1, pil_image_2):
|
263 |
"""Structured chest-X-ray report (streaming)."""
|
|
|
|
|
|
|
|
|
264 |
streamer = TextIteratorStreamer(chex_tok, skip_prompt=True, skip_special_tokens=True)
|
265 |
paths = []
|
266 |
for im in [pil_image_1, pil_image_2]:
|
@@ -270,6 +377,10 @@ def response_report_generation(pil_image_1, pil_image_2):
|
|
270 |
im.save(tfile.name)
|
271 |
paths.append(tfile.name)
|
272 |
|
|
|
|
|
|
|
|
|
273 |
device = get_model_device(chex_model)
|
274 |
anatomies = [
|
275 |
"View",
|
@@ -343,10 +454,12 @@ def response_report_generation(pil_image_1, pil_image_2):
|
|
343 |
yield clean_text(partial)
|
344 |
yield clean_text(partial)
|
345 |
|
346 |
-
|
347 |
@torch.no_grad()
|
348 |
def response_phrase_grounding(pil_image, prompt_text):
|
349 |
"""Very simple visual-grounding placeholder."""
|
|
|
|
|
|
|
350 |
if pil_image is None:
|
351 |
return "Please upload an image.", None
|
352 |
|
@@ -376,60 +489,96 @@ def response_phrase_grounding(pil_image, prompt_text):
|
|
376 |
|
377 |
return resp, pil_image
|
378 |
|
379 |
-
|
380 |
# =============================================================================
|
381 |
# Gradio UI
|
382 |
# =============================================================================
|
383 |
-
|
384 |
-
|
385 |
-
|
386 |
-
|
387 |
-
|
388 |
-
|
389 |
-
|
390 |
-
|
391 |
-
|
392 |
-
|
393 |
-
|
394 |
-
q_out = gr.Textbox(label="Answer")
|
395 |
-
q_btn.click(fn=med_agent.run, inputs=[q_in, q_img], outputs=q_out)
|
396 |
-
|
397 |
-
# ---------------------------------------------------------
|
398 |
-
with gr.Tab("Automatic masking (SAM-2)"):
|
399 |
-
seg_img = gr.Image(label="Image", type="pil")
|
400 |
-
seg_btn = gr.Button("Run segmentation")
|
401 |
-
seg_out = gr.Image(label="Overlay", type="pil")
|
402 |
-
seg_status = gr.Textbox(label="Status", interactive=False)
|
403 |
-
seg_btn.click(
|
404 |
-
fn=tumor_segmentation_interface,
|
405 |
-
inputs=seg_img,
|
406 |
-
outputs=[seg_out, seg_status],
|
407 |
-
)
|
408 |
|
409 |
-
|
410 |
-
|
411 |
-
|
412 |
-
|
413 |
-
|
414 |
-
|
415 |
-
|
416 |
-
|
417 |
-
|
418 |
-
|
419 |
-
|
420 |
-
|
421 |
-
|
422 |
-
|
423 |
-
|
424 |
-
|
425 |
-
|
426 |
-
|
427 |
-
|
428 |
-
|
429 |
-
|
430 |
-
|
431 |
-
|
432 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
433 |
|
434 |
if __name__ == "__main__":
|
|
|
435 |
demo.launch(server_name="0.0.0.0", server_port=7860, share=True)
|
|
|
6 |
|
7 |
β Changes β
|
8 |
-----------
|
9 |
+
1. Fixed SAM-2 installation and import issues
|
10 |
+
2. Added proper error handling for missing dependencies
|
11 |
+
3. Made SAM-2 functionality optional with graceful fallback
|
12 |
+
4. Added installation instructions and requirements check
|
13 |
"""
|
14 |
|
15 |
# ---------------------------------------------------------------------
|
16 |
# Standard libs
|
17 |
# ---------------------------------------------------------------------
|
|
|
|
|
|
|
|
|
|
|
18 |
import os
|
19 |
import sys
|
20 |
import uuid
|
21 |
import tempfile
|
22 |
+
import subprocess
|
23 |
+
import warnings
|
24 |
from threading import Thread
|
25 |
|
26 |
+
# Environment setup
|
27 |
+
os.environ["PYTORCH_ENABLE_MPS_FALLBACK"] = "1"
|
28 |
+
warnings.filterwarnings("ignore", message=r".*upsample_bicubic2d.*")
|
29 |
+
|
30 |
# ---------------------------------------------------------------------
|
31 |
# Third-party libs
|
|
|
|
|
32 |
# ---------------------------------------------------------------------
|
33 |
import torch
|
34 |
import numpy as np
|
35 |
from PIL import Image, ImageDraw
|
36 |
import gradio as gr
|
37 |
|
|
|
|
|
|
|
38 |
# =============================================================================
|
39 |
+
# Dependency checker and installer
|
40 |
# =============================================================================
|
41 |
+
def check_and_install_sam2():
|
42 |
+
"""Check if SAM-2 is available and attempt installation if needed."""
|
43 |
+
try:
|
44 |
+
# Try importing SAM-2
|
45 |
+
from sam2.build_sam import build_sam2
|
46 |
+
from sam2.automatic_mask_generator import SAM2AutomaticMaskGenerator
|
47 |
+
return True, "SAM-2 already available"
|
48 |
+
except ImportError:
|
49 |
+
print("SAM-2 not found. Attempting to install...")
|
50 |
+
try:
|
51 |
+
# Clone SAM-2 repository
|
52 |
+
if not os.path.exists("segment-anything-2"):
|
53 |
+
subprocess.run([
|
54 |
+
"git", "clone",
|
55 |
+
"https://github.com/facebookresearch/segment-anything-2.git"
|
56 |
+
], check=True)
|
57 |
+
|
58 |
+
# Install SAM-2
|
59 |
+
original_dir = os.getcwd()
|
60 |
+
os.chdir("segment-anything-2")
|
61 |
+
subprocess.run([sys.executable, "-m", "pip", "install", "-e", "."], check=True)
|
62 |
+
os.chdir(original_dir)
|
63 |
+
|
64 |
+
# Add to Python path
|
65 |
+
sys.path.insert(0, os.path.abspath("segment-anything-2"))
|
66 |
+
|
67 |
+
# Try importing again
|
68 |
+
from sam2.build_sam import build_sam2
|
69 |
+
from sam2.automatic_mask_generator import SAM2AutomaticMaskGenerator
|
70 |
+
return True, "SAM-2 installed successfully"
|
71 |
+
|
72 |
+
except Exception as e:
|
73 |
+
print(f"Failed to install SAM-2: {e}")
|
74 |
+
return False, f"SAM-2 installation failed: {e}"
|
75 |
+
|
76 |
+
# Check SAM-2 availability
|
77 |
+
SAM2_AVAILABLE, SAM2_STATUS = check_and_install_sam2()
|
78 |
+
print(f"SAM-2 Status: {SAM2_STATUS}")
|
79 |
|
80 |
# =============================================================================
|
81 |
+
# SAM-2 imports (conditional)
|
82 |
# =============================================================================
|
83 |
+
if SAM2_AVAILABLE:
|
84 |
+
try:
|
85 |
+
from sam2.build_sam import build_sam2
|
86 |
+
from sam2.automatic_mask_generator import SAM2AutomaticMaskGenerator
|
87 |
+
from sam2.modeling.sam2_base import SAM2Base
|
88 |
+
from sam2.utils.misc import get_device_index
|
89 |
+
except ImportError as e:
|
90 |
+
print(f"SAM-2 import error: {e}")
|
91 |
+
SAM2_AVAILABLE = False
|
92 |
|
93 |
+
# =============================================================================
|
94 |
+
# Qwen-VLM imports & helper
|
95 |
+
# =============================================================================
|
96 |
+
from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor
|
97 |
+
from qwen_vl_utils import process_vision_info
|
98 |
|
99 |
# =============================================================================
|
100 |
# CheXagent imports
|
101 |
# =============================================================================
|
102 |
from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer
|
103 |
|
|
|
104 |
# ---------------------------------------------------------------------
|
105 |
# Devices
|
106 |
# ---------------------------------------------------------------------
|
|
|
111 |
return torch.device("mps")
|
112 |
return torch.device("cpu")
|
113 |
|
|
|
114 |
# =============================================================================
|
115 |
# Qwen-VLM model & agent
|
116 |
# =============================================================================
|
|
|
118 |
_qwen_processor = None
|
119 |
_qwen_device = None
|
120 |
|
|
|
121 |
def load_qwen_model_and_processor(hf_token=None):
|
122 |
global _qwen_model, _qwen_processor, _qwen_device
|
123 |
if _qwen_model is None:
|
|
|
140 |
)
|
141 |
return _qwen_model, _qwen_processor, _qwen_device
|
142 |
|
|
|
143 |
class MedicalVLMAgent:
|
144 |
"""Light wrapper around Qwen-VLM with an optional image."""
|
145 |
|
|
|
182 |
trimmed = out[0][inputs.input_ids.shape[1] :]
|
183 |
return self.processor.decode(trimmed, skip_special_tokens=True).strip()
|
184 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
185 |
# =============================================================================
|
186 |
+
# SAM-2 model + AutomaticMaskGenerator (conditional)
|
187 |
# =============================================================================
|
188 |
+
def download_sam2_checkpoint():
|
189 |
+
"""Download SAM-2 checkpoint if not present."""
|
190 |
+
checkpoint_dir = "checkpoints"
|
191 |
+
checkpoint_file = "sam2.1_hiera_large.pt"
|
192 |
+
checkpoint_path = os.path.join(checkpoint_dir, checkpoint_file)
|
193 |
+
|
194 |
+
if not os.path.exists(checkpoint_path):
|
195 |
+
os.makedirs(checkpoint_dir, exist_ok=True)
|
196 |
+
print("Downloading SAM-2 checkpoint...")
|
197 |
+
try:
|
198 |
+
import urllib.request
|
199 |
+
url = "https://dl.fbaipublicfiles.com/segment_anything_2/092824/sam2.1_hiera_large.pt"
|
200 |
+
urllib.request.urlretrieve(url, checkpoint_path)
|
201 |
+
print("SAM-2 checkpoint downloaded successfully")
|
202 |
+
except Exception as e:
|
203 |
+
print(f"Failed to download SAM-2 checkpoint: {e}")
|
204 |
+
return None
|
205 |
+
|
206 |
+
return checkpoint_path
|
207 |
|
208 |
def initialize_sam2():
|
209 |
+
"""Initialize SAM-2 model and mask generator."""
|
210 |
+
if not SAM2_AVAILABLE:
|
211 |
+
return None, None
|
212 |
+
|
213 |
+
try:
|
214 |
+
# Download checkpoint if needed
|
215 |
+
checkpoint_path = download_sam2_checkpoint()
|
216 |
+
if checkpoint_path is None:
|
217 |
+
return None, None
|
218 |
+
|
219 |
+
# Config path (you may need to adjust this)
|
220 |
+
config_path = "segment-anything-2/sam2/configs/sam2.1/sam2.1_hiera_l.yaml"
|
221 |
+
if not os.path.exists(config_path):
|
222 |
+
config_path = "configs/sam2.1/sam2.1_hiera_l.yaml"
|
223 |
+
|
224 |
+
device = get_device()
|
225 |
+
print(f"[SAM-2] building model on {device}")
|
226 |
+
|
227 |
+
sam2_model = build_sam2(
|
228 |
+
config_path,
|
229 |
+
checkpoint_path,
|
230 |
+
device=device,
|
231 |
+
apply_postprocessing=False,
|
232 |
+
)
|
|
|
233 |
|
234 |
+
mask_gen = SAM2AutomaticMaskGenerator(
|
235 |
+
model=sam2_model,
|
236 |
+
points_per_side=32,
|
237 |
+
pred_iou_thresh=0.86,
|
238 |
+
stability_score_thresh=0.92,
|
239 |
+
crop_n_layers=0,
|
240 |
+
)
|
241 |
+
return sam2_model, mask_gen
|
242 |
+
|
243 |
+
except Exception as e:
|
244 |
+
print(f"[SAM-2] Failed to initialize: {e}")
|
245 |
+
return None, None
|
246 |
|
247 |
+
# Initialize SAM-2 (conditional)
|
248 |
+
_sam2_model, _mask_generator = None, None
|
249 |
+
if SAM2_AVAILABLE:
|
250 |
_sam2_model, _mask_generator = initialize_sam2()
|
251 |
+
if _sam2_model is not None:
|
252 |
+
print("[SAM-2] Successfully initialized!")
|
253 |
+
else:
|
254 |
+
print("[SAM-2] Initialization failed")
|
255 |
|
256 |
def automatic_mask_overlay(image_np: np.ndarray) -> np.ndarray:
|
257 |
"""Generate masks and alpha-blend them on top of the original image."""
|
|
|
274 |
return overlay
|
275 |
|
276 |
def tumor_segmentation_interface(image: Image.Image | None):
|
277 |
+
"""Tumor segmentation interface with proper error handling."""
|
278 |
if image is None:
|
279 |
return None, "Please upload an image."
|
280 |
|
281 |
+
if not SAM2_AVAILABLE:
|
282 |
+
return None, "SAM-2 is not available. Please check installation."
|
283 |
+
|
284 |
if _mask_generator is None:
|
285 |
return None, "SAM-2 not properly initialized. Check the console for errors."
|
286 |
|
|
|
293 |
return None, f"SAM-2 error: {e}"
|
294 |
|
295 |
# =============================================================================
|
296 |
+
# Simple fallback segmentation (when SAM-2 is not available)
|
297 |
# =============================================================================
|
298 |
+
def simple_segmentation_fallback(image: Image.Image | None):
|
299 |
+
"""Simple fallback segmentation using basic image processing."""
|
300 |
+
if image is None:
|
301 |
+
return None, "Please upload an image."
|
302 |
+
|
303 |
+
try:
|
304 |
+
import cv2
|
305 |
+
from skimage import segmentation, color
|
306 |
+
|
307 |
+
# Convert to numpy array
|
308 |
+
img_np = np.array(image.convert("RGB"))
|
309 |
+
|
310 |
+
# Simple watershed segmentation
|
311 |
+
gray = cv2.cvtColor(img_np, cv2.COLOR_RGB2GRAY)
|
312 |
+
_, binary = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY_INV + cv2.THRESH_OTSU)
|
313 |
+
|
314 |
+
# Remove noise
|
315 |
+
kernel = np.ones((3,3), np.uint8)
|
316 |
+
opening = cv2.morphologyEx(binary, cv2.MORPH_OPEN, kernel, iterations=2)
|
317 |
+
|
318 |
+
# Sure background area
|
319 |
+
sure_bg = cv2.dilate(opening, kernel, iterations=3)
|
320 |
+
|
321 |
+
# Finding sure foreground area
|
322 |
+
dist_transform = cv2.distanceTransform(opening, cv2.DIST_L2, 5)
|
323 |
+
_, sure_fg = cv2.threshold(dist_transform, 0.7*dist_transform.max(), 255, 0)
|
324 |
+
|
325 |
+
# Create overlay
|
326 |
+
overlay = img_np.copy()
|
327 |
+
overlay[sure_fg > 0] = [255, 0, 0] # Red overlay
|
328 |
+
|
329 |
+
# Alpha blend
|
330 |
+
result = cv2.addWeighted(img_np, 0.7, overlay, 0.3, 0)
|
331 |
+
|
332 |
+
return Image.fromarray(result), "Simple segmentation applied (SAM-2 not available)"
|
333 |
+
|
334 |
+
except Exception as e:
|
335 |
+
return None, f"Fallback segmentation error: {e}"
|
336 |
|
337 |
+
# =============================================================================
|
338 |
+
# CheXagent set-up
|
339 |
+
# =============================================================================
|
340 |
+
try:
|
341 |
+
chex_name = "StanfordAIMI/CheXagent-2-3b"
|
342 |
+
chex_tok = AutoTokenizer.from_pretrained(chex_name, trust_remote_code=True)
|
343 |
+
chex_model = AutoModelForCausalLM.from_pretrained(
|
344 |
+
chex_name, device_map="auto", trust_remote_code=True
|
345 |
+
)
|
346 |
+
chex_model = chex_model.half() if torch.cuda.is_available() else chex_model.float()
|
347 |
+
chex_model.eval()
|
348 |
+
CHEXAGENT_AVAILABLE = True
|
349 |
+
except Exception as e:
|
350 |
+
print(f"CheXagent not available: {e}")
|
351 |
+
CHEXAGENT_AVAILABLE = False
|
352 |
+
chex_tok, chex_model = None, None
|
353 |
|
354 |
def get_model_device(model):
|
355 |
+
if model is None:
|
356 |
+
return torch.device("cpu")
|
357 |
for p in model.parameters():
|
358 |
return p.device
|
359 |
return torch.device("cpu")
|
360 |
|
|
|
361 |
def clean_text(text):
|
362 |
return text.replace("</s>", "")
|
363 |
|
|
|
364 |
@torch.no_grad()
|
365 |
def response_report_generation(pil_image_1, pil_image_2):
|
366 |
"""Structured chest-X-ray report (streaming)."""
|
367 |
+
if not CHEXAGENT_AVAILABLE:
|
368 |
+
yield "CheXagent is not available. Please check installation."
|
369 |
+
return
|
370 |
+
|
371 |
streamer = TextIteratorStreamer(chex_tok, skip_prompt=True, skip_special_tokens=True)
|
372 |
paths = []
|
373 |
for im in [pil_image_1, pil_image_2]:
|
|
|
377 |
im.save(tfile.name)
|
378 |
paths.append(tfile.name)
|
379 |
|
380 |
+
if not paths:
|
381 |
+
yield "Please upload at least one image."
|
382 |
+
return
|
383 |
+
|
384 |
device = get_model_device(chex_model)
|
385 |
anatomies = [
|
386 |
"View",
|
|
|
454 |
yield clean_text(partial)
|
455 |
yield clean_text(partial)
|
456 |
|
|
|
457 |
@torch.no_grad()
|
458 |
def response_phrase_grounding(pil_image, prompt_text):
|
459 |
"""Very simple visual-grounding placeholder."""
|
460 |
+
if not CHEXAGENT_AVAILABLE:
|
461 |
+
return "CheXagent is not available. Please check installation.", None
|
462 |
+
|
463 |
if pil_image is None:
|
464 |
return "Please upload an image.", None
|
465 |
|
|
|
489 |
|
490 |
return resp, pil_image
|
491 |
|
|
|
492 |
# =============================================================================
|
493 |
# Gradio UI
|
494 |
# =============================================================================
|
495 |
+
def create_ui():
|
496 |
+
"""Create the Gradio interface."""
|
497 |
+
# Load Qwen model
|
498 |
+
try:
|
499 |
+
qwen_model, qwen_proc, qwen_dev = load_qwen_model_and_processor()
|
500 |
+
med_agent = MedicalVLMAgent(qwen_model, qwen_proc, qwen_dev)
|
501 |
+
qwen_available = True
|
502 |
+
except Exception as e:
|
503 |
+
print(f"Qwen model not available: {e}")
|
504 |
+
qwen_available = False
|
505 |
+
med_agent = None
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
506 |
|
507 |
+
with gr.Blocks(title="Medical AI Assistant") as demo:
|
508 |
+
gr.Markdown("# Combined Medical Q&A Β· SAM-2 Automatic Masking Β· CheXagent")
|
509 |
+
|
510 |
+
# Status information
|
511 |
+
with gr.Row():
|
512 |
+
gr.Markdown(f"""
|
513 |
+
**System Status:**
|
514 |
+
- Qwen VLM: {'β
Available' if qwen_available else 'β Not Available'}
|
515 |
+
- SAM-2: {'β
Available' if SAM2_AVAILABLE else 'β Not Available'}
|
516 |
+
- CheXagent: {'β
Available' if CHEXAGENT_AVAILABLE else 'β Not Available'}
|
517 |
+
""")
|
518 |
+
|
519 |
+
# Medical Q&A Tab
|
520 |
+
with gr.Tab("Medical Q&A"):
|
521 |
+
if qwen_available:
|
522 |
+
q_in = gr.Textbox(label="Question / description", lines=3)
|
523 |
+
q_img = gr.Image(label="Optional image", type="pil")
|
524 |
+
q_btn = gr.Button("Submit")
|
525 |
+
q_out = gr.Textbox(label="Answer")
|
526 |
+
q_btn.click(fn=med_agent.run, inputs=[q_in, q_img], outputs=q_out)
|
527 |
+
else:
|
528 |
+
gr.Markdown("β Medical Q&A is not available. Qwen model failed to load.")
|
529 |
+
|
530 |
+
# Segmentation Tab
|
531 |
+
with gr.Tab("Automatic masking"):
|
532 |
+
seg_img = gr.Image(label="Upload medical image", type="pil")
|
533 |
+
seg_btn = gr.Button("Run segmentation")
|
534 |
+
seg_out = gr.Image(label="Segmentation result", type="pil")
|
535 |
+
seg_status = gr.Textbox(label="Status", interactive=False)
|
536 |
+
|
537 |
+
if SAM2_AVAILABLE and _mask_generator is not None:
|
538 |
+
seg_btn.click(
|
539 |
+
fn=tumor_segmentation_interface,
|
540 |
+
inputs=seg_img,
|
541 |
+
outputs=[seg_out, seg_status],
|
542 |
+
)
|
543 |
+
else:
|
544 |
+
seg_btn.click(
|
545 |
+
fn=simple_segmentation_fallback,
|
546 |
+
inputs=seg_img,
|
547 |
+
outputs=[seg_out, seg_status],
|
548 |
+
)
|
549 |
+
|
550 |
+
# CheXagent Tabs
|
551 |
+
with gr.Tab("CheXagent β Structured report"):
|
552 |
+
if CHEXAGENT_AVAILABLE:
|
553 |
+
gr.Markdown("Upload one or two chest X-ray images; the report streams live.")
|
554 |
+
cx1 = gr.Image(label="Image 1", image_mode="L", type="pil")
|
555 |
+
cx2 = gr.Image(label="Image 2", image_mode="L", type="pil")
|
556 |
+
cx_report = gr.Markdown()
|
557 |
+
gr.Interface(
|
558 |
+
fn=response_report_generation,
|
559 |
+
inputs=[cx1, cx2],
|
560 |
+
outputs=cx_report,
|
561 |
+
live=True,
|
562 |
+
).render()
|
563 |
+
else:
|
564 |
+
gr.Markdown("β CheXagent structured report is not available.")
|
565 |
+
|
566 |
+
with gr.Tab("CheXagent β Visual grounding"):
|
567 |
+
if CHEXAGENT_AVAILABLE:
|
568 |
+
vg_img = gr.Image(image_mode="L", type="pil")
|
569 |
+
vg_prompt = gr.Textbox(value="Locate the highlighted finding:")
|
570 |
+
vg_text = gr.Markdown()
|
571 |
+
vg_out_img = gr.Image()
|
572 |
+
gr.Interface(
|
573 |
+
fn=response_phrase_grounding,
|
574 |
+
inputs=[vg_img, vg_prompt],
|
575 |
+
outputs=[vg_text, vg_out_img],
|
576 |
+
).render()
|
577 |
+
else:
|
578 |
+
gr.Markdown("β CheXagent visual grounding is not available.")
|
579 |
+
|
580 |
+
return demo
|
581 |
|
582 |
if __name__ == "__main__":
|
583 |
+
demo = create_ui()
|
584 |
demo.launch(server_name="0.0.0.0", server_port=7860, share=True)
|