split_shots.py

import os
import json
from pathlib import Path
from PIL import Image

data_dir = Path("data/animations")
output_dir = Path("data/shots")

# create output directory if it doesn't exist
output_dir.mkdir(parents=True, exist_ok=True)

# Find all .webp files in data/animations
webp_files = sorted(data_dir.glob("sample-*.webp"))

for webp_file in webp_files:
    # Extract NNN from filename
    base_name = webp_file.stem  # sample-NNN
    nnn = base_name.split("-")[1]
    hf_json_path = data_dir / f"sample-{nnn}.hf.json"
    if not hf_json_path.exists():
        print(f"Missing {hf_json_path}, skipping {webp_file}")
        continue

    # Load human feedback json
    with open(hf_json_path, "r") as f:
        hf_data = json.load(f)

    # Open the .webp image
    img = Image.open(webp_file)

    # Get scene change indices
    indices = hf_data.get("scene_change_indices", [])
    if not indices:
        print(f"No scene_change_indices in {hf_json_path}, skipping.")
        continue

    # Build split frame ranges
    ranges = []
    prev = 0
    for idx in indices:
        ranges.append((prev, idx))
        prev = idx + 1
    # Add final range to end of frames
    total_frames = getattr(img, 'n_frames', 1)
    ranges.append((prev, total_frames - 1))

    for i, (start, end) in enumerate(ranges):
        frames = []
        durations = []
        for frame_idx in range(start, end + 1):
            img.seek(frame_idx)
            frames.append(img.copy())
            # Get duration for each frame
            duration = img.info.get('duration', 40)  # default to 40ms if missing
            durations.append(duration)
        if frames:
            out_path = output_dir / f"sample-{nnn}-{i}.webp"
            if len(frames) == 1:
                frames[0].save(out_path, format="WEBP", duration=durations[0])
            else:
                frames[0].save(out_path, format="WEBP", save_all=True, append_images=frames[1:], duration=durations)
            print(f"Saved {out_path} ({len(frames)} frames)")