Upload 41 files

README.md

@@ -5,10 +5,13 @@ colorFrom: blue
 colorTo: red
 sdk: gradio
 sdk_version: 4.44.1
-app_file: app.py
-pinned: false
-license: mit
+app_file: app.py
+pinned: false
+license: mit
 short_description: Gaze-guided Gaussian-noise image regeneration
+hf_oauth: true
+hf_oauth_scopes:
+  - gated-repos
 ---
 
 # GazeCorrect — Gaze-Guided Regeneration

app.py

@@ -134,18 +134,23 @@ def device_dtype():
     return ("cuda", torch.float16) if torch.cuda.is_available() else ("cpu", torch.float32)
 
 
-@lru_cache(maxsize=1)
-def pipeline():
+@lru_cache(maxsize=2)
+def pipeline(token):
     from diffusers import StableDiffusionImg2ImgPipeline
     device, dtype = device_dtype()
-    token = os.environ.get("HF_TOKEN") or os.environ.get("HUGGINGFACEHUB_API_TOKEN")
     return StableDiffusionImg2ImgPipeline.from_pretrained(MODEL_ID, torch_dtype=dtype, token=token).to(device)
 
 
 @spaces.GPU
-def generate(image, points, description, spread, degree, feather, strength, steps, seed, progress=gr.Progress()):
+def generate(image, points, description, spread, degree, feather, strength, steps, seed,
+             oauth_token: gr.OAuthToken | None = None, progress=gr.Progress()):
     if image is None: return None, None, None, "Upload an image first."
     if not points: return None, None, None, "Add clicked or CSV gaze points first."
+    token = oauth_token.token if oauth_token is not None else (
+        os.environ.get("HF_TOKEN") or os.environ.get("HUGGINGFACEHUB_API_TOKEN")
+    )
+    if not token:
+        return None, None, None, "Sign in with Hugging Face first, then click Generate."
     seed = None if seed < 0 else int(seed)
     try:
         progress(.1, desc="Creating gaze attention")
@@ -154,7 +159,7 @@ def generate(image, points, description, spread, degree, feather, strength, step
         noised, mask = noisy_image(image, heat, degree, feather, seed)
         progress(.4, desc="Regenerating with RoentGen-v2")
         device, _ = device_dtype(); generator = None if seed is None else torch.Generator(device=device).manual_seed(seed)
-        output = pipeline()(prompt=description.strip() or "Normal chest radiograph.", image=noised,
+        output = pipeline(token)(prompt=description.strip() or "Normal chest radiograph.", image=noised,
                             strength=float(strength), guidance_scale=3.5,
                             num_inference_steps=int(steps), generator=generator).images[0].convert("RGB")
         corrected = Image.composite(output.resize(image.size), image.convert("RGB"), mask)
@@ -166,6 +171,7 @@ def generate(image, points, description, spread, degree, feather, strength, step
 with gr.Blocks(title="GazeCorrect") as demo:
     gr.Markdown("# GazeCorrect\nImage + gaze clicks/CSV + disease description → attention noise → corrected regenerated image. Use chest X-rays only; research use only.")
     image_state, points_state = gr.State(None), gr.State([])
+    gr.LoginButton("Sign in with Hugging Face")
     with gr.Row():
         with gr.Column():
             upload = gr.File(label="1. Upload chest X-ray", file_types=[".png", ".jpg", ".jpeg", ".webp", ".bmp"])

requirements.txt

@@ -1,4 +1,4 @@
-gradio==4.44.1
+gradio[oauth]==4.44.1
 torch>=2.1
 numpy
 pandas