Update app.py

app.py

@@ -1,3 +1,4 @@
+# app.py
 import os
 import torch
 import torch.nn.functional as F
@@ -5,59 +6,45 @@ import gradio as gr
 import numpy as np
 from PIL import Image, ImageDraw
 import torchvision.transforms.functional as TF
-from matplotlib import colaps
+from matplotlib import colormaps
 from transformers import AutoModel
 
 # ----------------------------
 # Configuration
 # ----------------------------
-# ⭐ Define available models, with the smaller one as default
-MODELS = {
-    "DINOv3 ViT-S+ (Small, Default)": "facebook/dinov3-vits16plus-pretrain-lvd1689m",
-    "DINOv3 ViT-H+ (Huge)": "facebook/dinov3-vith16plus-pretrain-lvd1689m",
-}
-DEFAULT_MODEL_NAME = "DINOv3 ViT-S+ (Small, Default)"
-
+MODEL_ID = "facebook/dinov3-vith16plus-pretrain-lvd1689m"
 PATCH_SIZE = 16
 DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
 
-# Normalization constants (standard for ImageNet)
 IMAGENET_MEAN = (0.485, 0.456, 0.406)
 IMAGENET_STD = (0.229, 0.224, 0.225)
 
-# ⭐ Cache for loaded models to avoid re-downloading
-model_cache = {}
-
 # ----------------------------
 # Model Loading (Hugging Face Hub)
 # ----------------------------
-def load_model_from_hub(model_id: str):
-    """Loads a DINOv3 model from the Hugging Face Hub."""
-    print(f"Loading model '{model_id}' from Hugging Face Hub...")
+def load_model_from_hub():
+    """Loads the DINOv3 model from the Hugging Face Hub."""
+    print(f"Loading model '{MODEL_ID}' from Hugging Face Hub...")
     try:
         token = os.environ.get("HF_TOKEN")
-        model = AutoModel.from_pretrained(model_id, token=token, trust_remote_code=True)
+        model = AutoModel.from_pretrained(MODEL_ID, token=token, trust_remote_code=True)
         model.to(DEVICE).eval()
         print(f"✅ Model loaded successfully on device: {DEVICE}")
         return model
     except Exception as e:
         print(f"❌ Failed to load model: {e}")
         raise gr.Error(
-            f"Could not load model '{model_id}'. "
+            f"Could not load model '{MODEL_ID}'. "
             "This is a gated model. Please ensure you have accepted the terms on its Hugging Face page "
             "and set your HF_TOKEN as a secret in your Space settings. "
             f"Original error: {e}"
         )
 
-def get_model(model_name: str):
-    """Gets a model from the cache or loads it if not present."""
-    model_id = MODELS[model_name]
-    if model_id not in model_cache:
-        model_cache[model_id] = load_model_from_hub(model_id)
-    return model_cache[model_id]
+# Load the model globally when the app starts
+model = load_model_from_hub()
 
 # ----------------------------
-# Helper Functions (resize, viz) - No changes here
+# Helper Functions (resize, viz)
 # ----------------------------
 def resize_to_grid(img: Image.Image, long_side: int, patch: int) -> torch.Tensor:
     w, h = img.size
@@ -77,10 +64,7 @@ def colorize(sim_map_up: np.ndarray, cmap_name: str = "viridis") -> Image.Image:
 def blend(base: Image.Image, heat: Image.Image, alpha: float = 0.55) -> Image.Image:
     base = base.convert("RGBA")
     heat = heat.convert("RGBA")
-    a = Image.new("L", heat.size, int(255 * alpha))
-    heat.putalpha(a)
-    out = Image.alpha_composite(base, heat)
-    return out.convert("RGB")
+    return Image.blend(base, heat, alpha=alpha)
 
 def draw_crosshair(img: Image.Image, x: int, y: int, radius: int = None) -> Image.Image:
     r = radius if radius is not None else max(2, PATCH_SIZE // 2)
@@ -112,31 +96,26 @@ def patch_neighborhood_box(r: int, c: int, Hp: int, Wp: int, rad: int, patch: in
     return (x0, y0, x1, y1)
 
 # ----------------------------
-# Feature Extraction (using transformers)
+# Feature Extraction
 # ----------------------------
 @torch.inference_mode()
-# ⭐ Pass the model object as an argument
-def extract_image_features(model, image_pil: Image.Image, target_long_side: int):
-    """
-    Extracts patch features from an image using the loaded Hugging Face model.
-    """
+def extract_image_features(image_pil: Image.Image, target_long_side: int):
     t = resize_to_grid(image_pil, target_long_side, PATCH_SIZE)
     t_norm = TF.normalize(t, IMAGENET_MEAN, IMAGENET_STD).unsqueeze(0).to(DEVICE)
     _, _, H, W = t_norm.shape
     Hp, Wp = H // PATCH_SIZE, W // PATCH_SIZE
-
+    
     outputs = model(t_norm)
-
+    
     n_special_tokens = 5
     patch_embeddings = outputs.last_hidden_state.squeeze(0)[n_special_tokens:, :]
-
+    
     X = F.normalize(patch_embeddings, p=2, dim=-1)
-
     img_resized = TF.to_pil_image(t)
     return {"X": X, "Hp": Hp, "Wp": Wp, "img": img_resized}
 
 # ----------------------------
-# Similarity inside the same image - No changes here
+# Similarity Logic
 # ----------------------------
 def click_to_similarity_in_same_image(
     state: dict,
@@ -149,21 +128,17 @@ def click_to_similarity_in_same_image(
 ):
     if not state:
         return None, None, None, None
-
     X = state["X"]
     Hp, Wp = state["Hp"], state["Wp"]
     base_img = state["img"]
     img_w, img_h = base_img.size
-
     x_pix, y_pix = click_xy
     col = int(np.clip(x_pix // PATCH_SIZE, 0, Wp - 1))
     row = int(np.clip(y_pix // PATCH_SIZE, 0, Hp - 1))
     idx = row * Wp + col
-
     q = X[idx]
     sims = torch.matmul(X, q)
     sim_map = sims.view(Hp, Wp)
-
     if exclude_radius_patches > 0:
         rr, cc = torch.meshgrid(
             torch.arange(Hp, device=sims.device),
@@ -172,17 +147,14 @@ def click_to_similarity_in_same_image(
         )
         mask = (torch.abs(rr - row) <= exclude_radius_patches) & (torch.abs(cc - col) <= exclude_radius_patches)
         sim_map = sim_map.masked_fill(mask, float("-inf"))
-
     sim_up = F.interpolate(
         sim_map.unsqueeze(0).unsqueeze(0),
         size=(img_h, img_w),
         mode="bicubic",
         align_corners=False,
     ).squeeze().detach().cpu().numpy()
-
     heatmap_pil = colorize(sim_up, cmap_name)
     overlay_pil = blend(base_img, heatmap_pil, alpha=alpha)
-
     overlay_boxes_pil = overlay_pil
     if topk and topk > 0:
         flat = sim_map.view(-1)
@@ -199,92 +171,82 @@ def click_to_similarity_in_same_image(
                 for r, c in [divmod(j.item(), Wp) for j in top_idx]
             ]
             overlay_boxes_pil = draw_boxes(overlay_pil, boxes, outline="yellow", width=3, labels=True)
-
     marked_ref = draw_crosshair(base_img, x_pix, y_pix, radius=PATCH_SIZE // 2)
     return marked_ref, heatmap_pil, overlay_pil, overlay_boxes_pil
 
 # ----------------------------
 # Gradio UI
 # ----------------------------
-with gr.Blocks(theme=gr.themes.Soft(), title="DINOv3 Single-Image Patch Similarity") as demo:
-    gr.Markdown("# 🦖 DINOv3 Single-Image Patch Similarity")
-    gr.Markdown("## Running on CPU-only Space, feature extraction can take a moment")
-    gr.Markdown("1. **Choose a model**. 2. Upload an image. 3. Click **Process Image**. 4. **Click anywhere on the processed image** to find similar regions.")
-
+with gr.Blocks(theme=gr.themes.Soft(), title="DINOv3 Patch Similarity") as demo:
+    gr.Markdown("# 🦖 DINOv3: Visualizing Patch Similarity")
+    gr.Markdown(
+        "Upload an image, then **click anywhere** on it to find the most visually similar regions. "
+        "**Note:** If running on a CPU-only Space, feature extraction after uploading an image can take a moment."
+    )
+    
     app_state = gr.State()
-
+    
     with gr.Row():
-        with gr.Column(scale=1):
-            # ⭐ ADDED MODEL DROPDOWN
-            model_name_dd = gr.Dropdown(
-                label="1. Choose a Model",
-                choices=list(MODELS.keys()),
-                value=DEFAULT_MODEL_NAME,
-            )
+        with gr.Column(scale=2):
             input_image = gr.Image(
-                label="2. Upload Image",
+                label="Image (click anywhere)",
                 type="pil",
                 value="https://images.squarespace-cdn.com/content/v1/607f89e638219e13eee71b1e/1684821560422-SD5V37BAG28BURTLIXUQ/michael-sum-LEpfefQf4rU-unsplash.jpg"
             )
-            target_long_side = gr.Slider(
-                minimum=224, maximum=1024, value=768, step=16,
-                label="Processing Resolution",
-                info="Higher values = more detail but slower processing",
-            )
-            process_button = gr.Button("3. Process Image", variant="primary")
-
-            with gr.Row():
-                alpha = gr.Slider(0.0, 1.0, value=0.55, step=0.05, label="Overlay opacity")
+            with gr.Accordion("⚙️ Visualization Controls", open=True):
+                target_long_side = gr.Slider(
+                    minimum=224, maximum=1024, value=768, step=16,
+                    label="Processing Resolution",
+                    info="Higher values = more detail but slower processing",
+                )
+                alpha = gr.Slider(0.0, 1.0, value=0.55, step=0.05, label="Overlay Opacity")
                 cmap = gr.Dropdown(
                     ["viridis", "magma", "plasma", "inferno", "turbo", "cividis"],
-                    value="viridis", label="Colormap",
+                    value="viridis", label="Heatmap Colormap",
                 )
-        with gr.Column(scale=1):
-            exclude_r = gr.Slider(0, 10, value=0, step=1, label="Exclude radius (patches)")
-            topk = gr.Slider(0, 200, value=20, step=1, label="Top-K boxes")
-            box_radius = gr.Slider(0, 10, value=1, step=1, label="Box radius (patches)")
-
-    with gr.Row():
-        marked_image = gr.Image(label="4. Click on this image", interactive=True)
-        heatmap_output = gr.Image(label="Similarity heatmap", interactive=False)
-    with gr.Row():
-        overlay_output = gr.Image(label="Overlay (image ⊕ heatmap)", interactive=False)
-        overlay_boxes_output = gr.Image(label="Overlay + top-K similar patch boxes", interactive=False)
-
-    # ⭐ UPDATED to take model_name as input
-    def _process_image(model_name: str, img: Image.Image, long_side: int, progress=gr.Progress(track_tqdm=True)):
+            with gr.Accordion("⚙️ Similarity Controls", open=True):
+                exclude_r = gr.Slider(0, 10, value=0, step=1, label="Exclude Radius (patches)", info="Ignore patches around the click point.")
+                topk = gr.Slider(0, 50, value=10, step=1, label="Top-K Boxes", info="Number of similar regions to highlight.")
+                box_radius = gr.Slider(0, 10, value=1, step=1, label="Box Radius (patches)", info="Size of the highlight box.")
+
+        with gr.Column(scale=3):
+            marked_image = gr.Image(label="Your Click (on processed image)", interactive=False)
+            with gr.Tabs():
+                with gr.TabItem("📦 Bounding Boxes"):
+                    overlay_boxes_output = gr.Image(label="Overlay + Top-K Similar Patches", interactive=False)
+                with gr.TabItem("🔥 Heatmap"):
+                    heatmap_output = gr.Image(label="Similarity Heatmap", interactive=False)
+                with gr.TabItem(" blended"):
+                    overlay_output = gr.Image(label="Blended Overlay (Image + Heatmap)", interactive=False)
+
+    def _on_upload_or_slider_change(img: Image.Image, long_side: int, progress=gr.Progress(track_tqdm=True)):
         if img is None:
-            gr.Warning("Please upload an image first!")
             return None, None
-        
-        progress(0, desc=f"Loading model '{model_name}'...")
-        model = get_model(model_name)
-        
-        progress(0.5, desc="Extracting features...")
-        st = extract_image_features(model, img, int(long_side))
-        
-        progress(1, desc="Done! You can now click on the image.")
-        return st["img"], st
+        progress(0, desc="🦖 Extracting DINOv3 features...")
+        st = extract_image_features(img, int(long_side))
+        progress(1, desc="✅ Done!")
+        # Clear old results when a new image is uploaded
+        return st["img"], st, None, None, None, None
 
     def _on_click(st, a: float, m: str, excl: int, k: int, box_rad: int, evt: gr.SelectData):
         if not st or evt is None:
-            gr.Warning("Please process an image before clicking on it.")
-            return None, None, None, None
-        return click_to_similarity_in_same_image(
+            # Return current state if no click data
+            return st.get("img"), None, None, None
+        
+        marked, heat, overlay, boxes = click_to_similarity_in_same_image(
             st, click_xy=evt.index, exclude_radius_patches=int(excl),
             topk=int(k), alpha=float(a), cmap_name=m,
             box_radius_patches=int(box_rad),
         )
+        return marked, heat, overlay, boxes
 
-    # ⭐ UPDATED EVENT WIRING to include the dropdown
-    inputs_for_processing = [model_name_dd, input_image, target_long_side]
-    outputs_for_processing = [marked_image, app_state]
+    # Wire events
+    inputs_for_update = [input_image, target_long_side]
+    outputs_for_upload = [marked_image, app_state, heatmap_output, overlay_output, overlay_boxes_output, marked_image]
 
-    process_button.click(
-        _process_image,
-        inputs=inputs_for_processing,
-        outputs=outputs_for_processing
-    )
+    input_image.upload(_on_upload_or_slider_change, inputs=inputs_for_update, outputs=outputs_for_upload)
+    target_long_side.change(_on_upload_or_slider_change, inputs=inputs_for_update, outputs=outputs_for_upload)
+    demo.load(_on_upload_or_slider_change, inputs=inputs_for_update, outputs=outputs_for_upload)
 
     marked_image.select(
         _on_click,