|
Download evaluation/examples/INPUT_FORMAT.md from NoizAI/Helixbench: direct link, hf CLI and curl.
- Browser
- Download file 573 Bytes
-
https://huggingface.co/datasets/NoizAI/Helixbench/resolve/main/evaluation/examples/INPUT_FORMAT.md
- Command line
-
hf download hf://datasets/NoizAI/Helixbench/evaluation/examples/INPUT_FORMAT.md
-
curl -L -o INPUT_FORMAT.md https://huggingface.co/datasets/NoizAI/Helixbench/resolve/main/evaluation/examples/INPUT_FORMAT.md
573 Bytes
Generated media manifest
helixbench prepare --generated accepts JSONL with exactly the frozen 1,015 IDs. Each row gives an absolute MP4 path and its SHA256:
{"sample_id":"YB0006_Indoor_navigation_0en_ZwGyokU","path":"/absolute/generated/YB0006_Indoor_navigation_0en_ZwGyokU.mp4","sha256":"<64 hex digits>"}
The packaged reference features supply GT for KL/FAD. Optionally pass --gt with the same JSONL shape to recompute GT features from exact GT MP4s. The frozen caption_av, event/perspective subsets, and first-frame SHA are read from the package.