loop detection

detect_loops.py

@@ -17,9 +17,16 @@ def extract_frames(webp_path):
     return frames
 
 def compute_sim(f1, f2):
-    # f1 and f2 are already grayscale and downscaled
-    mse = np.mean((f1.astype(np.float32) - f2.astype(np.float32)) ** 2)
-    return mse
+    assert f1.shape == f2.shape, f"Shape mismatch: {f1.shape} vs {f2.shape}"
+    assert f1.shape[2] == 3, f"Expected 3 channels, got {f1.shape[2]}"
+    assert f1.dtype == np.float32, f"Expected float32, got {f1.dtype}"
+    assert f2.dtype == np.float32, f"Expected float32, got {f2.dtype}"
+    # Flatten to 1D vectors
+    v1 = f1.flatten()
+    v2 = f2.flatten()
+    eps = 1e-8
+    cos_sim = np.dot(v1, v2) / (np.linalg.norm(v1) * np.linalg.norm(v2) + eps)
+    return cos_sim
 
 def detect_loops(frames, min_len=6, max_len=40, top_k=3):
     import time
@@ -48,11 +55,11 @@ def detect_loops(frames, min_len=6, max_len=40, top_k=3):
             # Compare composite frames directly
             start_comp = composite_frames[i].astype(np.float32)
             end_comp = composite_frames[j].astype(np.float32)
-            sim = compute_sim(start_comp, end_comp)
+            cos_sim = compute_sim(start_comp, end_comp)
             t1 = time.time()
-            score = -sim  # Lower MSE is better, so negate for sorting
-            print(f"Loop ({i},{j}): Composite MSE={sim:.1f} (t={t1-t0:.3f}s)")
-            candidates.append((score, i, j, sim))
+            score = cos_sim  # Higher cosine similarity is better
+            print(f"Loop ({i},{j}): Cosine similarity={cos_sim:.4f} (t={t1-t0:.3f}s)")
+            candidates.append((score, i, j, cos_sim))
     # Sort by score descending
     candidates.sort(reverse=True)
     return candidates[:top_k]
@@ -99,26 +106,26 @@ if __name__ == "__main__":
         # Save all candidates and their scores to JSON
         import json
         loop_json = []
-        for score, i, j, sim in loops:
+        for score, i, j, cos_sim in loops:
             loop_json.append({
                 "start": int(i),
                 "end": int(j),
                 "score": float(score),
-                "sim": float(sim)
+                "cos_sim": float(cos_sim)
             })
         json_name = f"{webp_path.stem}.loop.json"
         json_path = output_dir / json_name
         with open(json_path, "w") as f:
             json.dump(loop_json, f, indent=2)
         print(f"Saved loop candidates: {json_path}")
-        for idx, (score, i, j, sim) in enumerate(loops):
-            print(f"Loop candidate: start={i}, end={j}, score={score:.3f}, SIM={sim:.3f}")
+        for idx, (score, i, j, cos_sim) in enumerate(loops):
+            print(f"Loop candidate: start={i}, end={j}, score={score:.4f}, COS_SIM={cos_sim:.4f}")
             # Extract loop frames (seamless looping: frames[i:j])
             loop_frames = frames[i:j]
             # Convert BGR (OpenCV) to RGB for PIL
             pil_frames = [Image.fromarray(cv2.cvtColor(f, cv2.COLOR_BGR2RGB)) for f in loop_frames]
             # Save as animated webp
-            out_name = f"{webp_path.stem}.loop-{i}-{j}.webp"
+            out_name = f"{webp_path.stem}.loop-{idx}.webp"
             out_path = output_dir / out_name
             pil_frames[0].save(
                 out_path,