Fix SAM3 imports and API usage

app.py

@@ -47,17 +47,13 @@ def load_sam3():
         return SAM3_PREDICTOR
 
     import torch
-    from sam3 import SAM3Predictor
+    from sam3.model_builder import build_sam3_image_model
+    from sam3.model.sam3_image_processor import Sam3Processor
 
     print("Loading SAM3 model...")
 
-    device = "cuda" if torch.cuda.is_available() else "cpu"
-
-    SAM3_PREDICTOR = SAM3Predictor.from_pretrained(
-        "facebook/sam3-hiera-large",
-        device=device,
-        token=os.environ.get("HF_TOKEN")
-    )
+    model = build_sam3_image_model()
+    SAM3_PREDICTOR = Sam3Processor(model)
 
     print("✓ SAM3 loaded")
     return SAM3_PREDICTOR
@@ -96,24 +92,37 @@ def segment_with_text(image: np.ndarray, text_prompt: str):
         return None, None, "❌ No text prompt provided"
 
     try:
-        predictor = load_sam3()
+        from PIL import Image as PILImage
+        processor = load_sam3()
+
+        # Convert to PIL
+        if isinstance(image, np.ndarray):
+            pil_image = PILImage.fromarray(image)
+        else:
+            pil_image = image
 
         # Run SAM3 with text prompt
-        predictor.set_image(image)
-        masks, scores, _ = predictor.predict(text=text_prompt)
+        state = processor.set_image(pil_image)
+        output = processor.set_text_prompt(state=state, prompt=text_prompt)
+
+        if output is None or "masks" not in output:
+            return image, None, "⚠️ No object found"
 
-        if masks is None or len(masks) == 0:
+        masks = output["masks"]
+        scores = output.get("scores", [1.0])
+
+        if len(masks) == 0:
             return image, None, "⚠️ No object found"
 
         # Use best mask
-        best_idx = np.argmax(scores)
-        mask = masks[best_idx]
+        best_idx = np.argmax(scores) if len(scores) > 0 else 0
+        mask = np.array(masks[best_idx])
 
         # Create overlay
         overlay = image.copy()
-        overlay[mask] = (overlay[mask] * 0.5 + np.array([0, 255, 0]) * 0.5).astype(np.uint8)
+        overlay[mask > 0] = (overlay[mask > 0] * 0.5 + np.array([0, 255, 0]) * 0.5).astype(np.uint8)
 
-        return overlay, mask.astype(np.uint8) * 255, f"✓ Found: {text_prompt}"
+        return overlay, (mask > 0).astype(np.uint8) * 255, f"✓ Found: {text_prompt}"
 
     except Exception as e:
         import traceback
@@ -128,28 +137,40 @@ def segment_with_click(image: np.ndarray, evt: gr.SelectData):
         return None, None, "❌ No image provided"
 
     try:
-        predictor = load_sam3()
+        from PIL import Image as PILImage
+        processor = load_sam3()
+
+        # Convert to PIL
+        if isinstance(image, np.ndarray):
+            pil_image = PILImage.fromarray(image)
+        else:
+            pil_image = image
 
         # Get click coordinates
-        point = np.array([[evt.index[0], evt.index[1]]])
-        label = np.array([1])  # foreground
+        point = [evt.index[0], evt.index[1]]
 
-        predictor.set_image(image)
-        masks, scores, _ = predictor.predict(
-            point_coords=point,
-            point_labels=label,
-            multimask_output=True
-        )
+        # Run SAM3 with point prompt
+        state = processor.set_image(pil_image)
+        output = processor.set_point_prompt(state=state, points=[point], labels=[1])
+
+        if output is None or "masks" not in output:
+            return image, None, "⚠️ No object found"
+
+        masks = output["masks"]
+        scores = output.get("scores", [1.0])
+
+        if len(masks) == 0:
+            return image, None, "⚠️ No object found"
 
         # Use best mask
-        best_idx = np.argmax(scores)
-        mask = masks[best_idx]
+        best_idx = np.argmax(scores) if len(scores) > 0 else 0
+        mask = np.array(masks[best_idx])
 
         # Create overlay
         overlay = image.copy()
-        overlay[mask] = (overlay[mask] * 0.5 + np.array([0, 255, 0]) * 0.5).astype(np.uint8)
+        overlay[mask > 0] = (overlay[mask > 0] * 0.5 + np.array([0, 255, 0]) * 0.5).astype(np.uint8)
 
-        return overlay, mask.astype(np.uint8) * 255, "✓ Object selected"
+        return overlay, (mask > 0).astype(np.uint8) * 255, "✓ Object selected"
 
     except Exception as e:
         import traceback