|
Download README.md from Swindl/jevvideo-long-data: direct link, hf CLI and curl.
- Browser
- Download file 2.05 kB
-
https://huggingface.co/datasets/Swindl/jevvideo-long-data/resolve/main/README.md
- Command line
-
hf download hf://datasets/Swindl/jevvideo-long-data/README.md
-
curl -L -H "Authorization: Bearer $HF_TOKEN" -o README.md https://huggingface.co/datasets/Swindl/jevvideo-long-data/resolve/main/README.md
2.05 kB
| # Jevvideo long-video backup (private) | |
| Personal backup, not for redistribution. CG-Bench and Ego4D (LV-Haystack) videos stay under their own licenses. | |
| - `checkpoints/<TAG>/`: `runs/<TAG>_experiment.json` (settings), `runs/<TAG>_ps3_r2.json` (final policy set, absolute node paths), `ps3_r2.relative.json` (same, paths relative to this folder), and the LoRA snapshots it names under `runs/<run>/snapshots/step_XXXX/`. | |
| - `onejev_sft_videomme_mix_v1`: 长短混合 (code 6b4d95c) | |
| - `onejev_sft_videomme_mix_mseg_v1`: 32帧长短混合(新代码) (code 18cf7dd) | |
| - `onejev_sft_videomme_pad3_v1_f128`: 128帧正式 (code 6b4d95c + infra fixes) | |
| - `manifests/<TAG>/cgbench_lvh_dev_vts.jsonl`: the CG-Bench + LV-Haystack dev split each run used (`video_path` = the original Blob path). | |
| - `videos/cgbench/`: from `/blob/t-xinyangliu/data/cgbench/videos/`; `videos/lvhaystack_ego4d_full_scale/`: from `/blob/t-xinyangliu/data/lvhaystack/ego4d/v2/full_scale/` (the union over the three dev manifests). | |
| ## 2026-11-04: the full LV-Haystack train split (from ARC) | |
| - `videos/lvhaystack_ego4d_full_scale/`: every video of the LongVideoHaystack train split that ARC has, 857 of its | |
| 858 videos (516.9 GiB, Ego4D full scale; never downloaded: 0ca23a40-6daf-4503-bfa2-f315a79b7317). It covers | |
| the train and dev rows of every ARC run (they all come from this split), including the 5.2 h and 3.09 h videos the | |
| fused run drops. | |
| - `lvhaystack/`: ARC's LV-Haystack root without its videos (`scripts/prepare_lvhaystack.py`): | |
| - `annotations/data/{train,val,test}-00000-of-00001.parquet`: MLL-Lab/LongVideoHaystack @ 2ad05dee (train | |
| 12,892 questions on 858 videos). | |
| - `manifests/lvhaystack_{all,train,dev,test}.jsonl` and `lvhaystack_stats.json`: the prepared rows; train and dev both | |
| come from the LVH train split, test is LVH val (this upload adds none of its videos). `video_path` = the ARC path | |
| `/projects/swan_research_dlm/czwang26/jev_data/lvhaystack/ego4d/v2/full_scale/`. | |
| - `PREPARED.json`, `video_uids.txt`, `missing_video_uids.txt`. | |