How lc1471/ml-csd361 works
AVCAffe Preprocessing prepares clip-level audio/video data and labels for an Affect-Aware Dynamic UI project. The main workflow builds a manifest, converts audio and face crops into aligned two-second windows, then exposes normalized clips and labels through a reader or PyTorch wrapper. Separate utilities compute normalization statistics and check alignment or audio availability. The README gives little detail beyond the product name; unsampled crop-generation code and other utilities are represented cautiously, and edges are limited to documented or excerpted relationships.
Main entry points: dataset.py Clip reader · windows.py Aligned windows · audio_features.py Audio features · compute_stats.py Stats computation
Components 11
Dataset preparation
- Clip manifest
src/preprocessing/manifest.py - Crop preparation
src/preprocessing/add_crops.py
Feature extraction
- Audio features
src/preprocessing/audio_features.py - Video features
src/preprocessing/video_features.py - Aligned windows
src/preprocessing/windows.py
Dataset loading
- Clip reader
src/preprocessing/dataset.py - PyTorch adapter
src/preprocessing/torch_dataset.py
Stats and checks
- Normalization stats
configs/norm_stats.json - Stats computation
src/preprocessing/compute_stats.py - Alignment check
src/preprocessing/check_alignment.py - Audio check
src/preprocessing/check_no_audio.py
Connections 11
- Stats computation → Aligned windows: builds clips
src/preprocessing/compute_stats.py - Stats computation → Normalization stats: writes stats
src/preprocessing/compute_stats.py - Clip reader → Clip manifest: reads manifest
src/preprocessing/dataset.py - Clip reader → Normalization stats: reads stats
src/preprocessing/dataset.py - Clip reader → Aligned windows: builds clips
src/preprocessing/dataset.py - Aligned windows → Audio features: extracts audio
src/preprocessing/windows.py - Aligned windows → Video features: loads crops
src/preprocessing/windows.py - PyTorch adapter → Clip reader: wraps reader
src/preprocessing/torch_dataset.py - Alignment check → Audio features: checks audio
src/preprocessing/check_alignment.py - Alignment check → Video features: checks video
src/preprocessing/check_alignment.py - Audio check → Audio features: loads audio
src/preprocessing/check_no_audio.py