Datasets:

HuggingFriends
/

physion-eval-videoedit-100

Tasks:

Text-to-Video

Video-to-Video

Modalities:

Video

Size:

< 1K

Libraries:

Datasets

License:

Dataset card Data Studio Files Files and versions

xet

Community

physion-eval-videoedit-100 / README.md

HuggingFriends

Upload README.md with huggingface_hub

51365a4 verified 22 days ago

preview code

Raw History Blame Contribute Delete

1.32 kB

metadata

license: other
task_categories:
  - text-to-video
  - video-to-video
pretty_name: Physion-Eval VideoEdit 100

Physion-Eval VideoEdit 100

A 100-item working subset derived from PhysionLabs/Physion-Eval for physical-glitch video-edit experiments.

Each item is an annotated UNREALISTIC Physion-Eval video with a concise source-video-conditioned edit prompt. The original Physion provenance is preserved in manifest.jsonl.

Files:

manifest.jsonl: one row per video, including category, severity, source model, diagnosis, prompt, SHA256, and hf_video_url.
videos/: 100 individual MP4 files for direct model/API input.
review10.json / review10.md: 10 hand-review candidates with direct links.
review10_contact_sheets/: frame contact sheets for the 10 candidates.

Source provenance:

Source dataset: PhysionLabs/Physion-Eval
Local binary manifest used for this package: /mnt/bn/embodied-lf3/masiyuan/v2en_repro/eval802_scoring_prep/physion_eval_test880_20260917/physion_eval_binary_test880_manifest.jsonl
Full metadata used for reasoning/category: /mnt/bn/embodied-lf3/masiyuan/WorldJudge/data/judgebench/physion_eval/Physion_Eval_20260322.json

Prompt format intentionally stays short: fix the annotated physical glitch at the time hint, preserve camera/scene/unrelated objects.