- Detect CUDA version at venv creation time and install matching jax[cuda12/13]
instead of hardcoded jax[cuda13] — was broken on CUDA 12.x (most systems)
- Include fps in cache hash: same video+caption at different fps previously
returned stale cached features with wrong frame sampling
- Guard frame index lists with max(1,...)/max(8,...) to prevent torch.stack([])
crash on very short input clips; sync minimum is 8 to match Synchformer's
segment size requirement
- Remove mediapy from managed venv packages — not imported anywhere
- Warn when caption_cot is empty (produces degenerate text features)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Saves frames as uint8 .npy instead of H.264 MP4, eliminating the
lossy codec roundtrip. extract_features.py loads .npy directly and
skips decord when given a numpy file. Passes --source_fps for
correct temporal sampling.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Each step now prints elapsed seconds on completion.
Total time printed at the end to identify bottlenecks.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
T5-Gemma outputs BFloat16 which numpy does not support.
Cast all feature tensors with .float() before .numpy().
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Creates data_utils/v2a_utils/feature_utils_288.py with FeaturesUtils:
- T5-Gemma text encoding via transformers
- VideoPrism video encoding via JAX videoprism package
- Synchformer visual encoder loading from checkpoint
Also fixes extract_features.py to add plugin root to sys.path so
data_utils is importable in the subprocess venv.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Ethanfel