visualizing motion

data/shots/sample-000-0.motion.mp4

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d7b60d94b5ceea0e4009f493ace81406301e0572ba21ee4f5964d8a95d4b394
+size 265232

requirements.txt

@@ -3,3 +3,4 @@ torch
 numpy
 matplotlib
 streamlit
+opencv-python

visualize_motion.py

@@ -0,0 +1,84 @@
+import cv2
+import numpy as np
+from pathlib import Path
+from PIL import Image
+
+# Select files using glob (for now, only process the first file for testing)
+shots_dir = Path('data/shots')
+files = sorted(shots_dir.glob('sample-000-0.webp'))  # Change pattern to 'sample-*.webp' for batch
+if not files:
+    print('No files found.')
+    exit(1)
+
+# Process each file serially (for now, just one file)
+for webp_path in files:
+    print(f'Processing {webp_path}')
+    # Extract all frames from the animated webp using PIL
+    frames = []
+    frame_durations = []
+    with Image.open(webp_path) as im:
+        try:
+            while True:
+                frame = im.convert('RGB')
+                frames.append(np.array(frame)[:, :, ::-1])  # Convert RGB to BGR for OpenCV
+                # Get duration in ms for this frame (default to 100ms if not present)
+                duration = im.info.get('duration', 100)
+                frame_durations.append(duration)
+                im.seek(im.tell() + 1)
+        except EOFError:
+            pass
+
+    # Debug: check extracted frames
+    print(f"Extracted {len(frames)} frames from {webp_path}")
+    if len(frames) > 0:
+        print(f"First frame shape: {frames[0].shape}, dtype: {frames[0].dtype}, min: {frames[0].min()}, max: {frames[0].max()}")
+
+    # Compute dense optical flow and overlay visualization
+    hsv = None
+    motion_frames = []
+    for i in range(1, len(frames)):
+        prev = cv2.cvtColor(frames[i-1], cv2.COLOR_BGR2GRAY)
+        curr = cv2.cvtColor(frames[i], cv2.COLOR_BGR2GRAY)
+        flow = cv2.calcOpticalFlowFarneback(prev, curr, None, 0.5, 3, 15, 3, 5, 1.2, 0)
+        mag, ang = cv2.cartToPolar(flow[...,0], flow[...,1])
+        print(f"Frame {i}: flow mag min={mag.min()}, max={mag.max()}, mean={mag.mean()}")
+        if np.all(mag == 0):
+            print(f"Frame {i}: All zero motion, skipping.")
+            continue
+        # Draw flow as arrows on a grid
+        step = 16  # grid size
+        arrow_color = (0, 255, 0)  # green
+        arrow_thickness = 1
+        overlay = frames[i].copy()
+        h, w = prev.shape
+        for y in range(0, h, step):
+            for x in range(0, w, step):
+                fx, fy = flow[y, x]
+                end_x = int(x + fx * 4)
+                end_y = int(y + fy * 4)
+                cv2.arrowedLine(overlay, (x, y), (end_x, end_y), arrow_color, arrow_thickness, tipLength=0.3)
+        motion_frames.append(overlay)
+
+    # Save as mp4
+    if motion_frames:
+        height, width, _ = motion_frames[0].shape
+        # Calculate FPS from frame durations (use mean duration between frames)
+        if len(frame_durations) > 1:
+            # Use durations between frames (skip first frame)
+            mean_duration = np.mean(frame_durations[1:])
+        else:
+            mean_duration = 100
+        fps = 1000.0 / mean_duration if mean_duration > 0 else 10
+        print(f"Using FPS: {fps:.2f} (mean frame duration: {mean_duration} ms)")
+        if hasattr(cv2, 'VideoWriter_fourcc'):
+            fourcc = cv2.VideoWriter_fourcc(*'avc1')  # More compatible MP4 codec for macOS
+        else:
+            raise RuntimeError('cv2.VideoWriter_fourcc is not available in your OpenCV installation. Please update OpenCV.')
+        out_path = webp_path.parent / f"{webp_path.stem}.motion.mp4"
+        out = cv2.VideoWriter(str(out_path), fourcc, fps, (width, height))
+        for f in motion_frames:
+            out.write(f)
+        out.release()
+        print(f'Saved {out_path}')
+    else:
+        print('No motion frames to save for', webp_path)