VBVR-Pro-RL / README.md
junxiangjoe's picture
Align badges with scorer README
7f5a723 verified
|
Raw
History Blame Contribute Delete
7.95 kB
metadata
license: cc-by-nc-4.0
language:
  - en
tags:
  - video-generation
  - image-to-video
  - visual-reasoning
  - reinforcement-learning
pretty_name: VBVR-Pro-RL
size_categories:
  - 10K<n<100K
configs:
  - config_name: video
    default: true
    data_files:
      - split: train
        path: preview/video.parquet
  - config_name: image
    data_files:
      - split: train
        path: preview/image.parquet

VBVR-Pro-RL

The reinforcement-learning split of VBVR-Pro: 50 parameterized tasks × 1,000 instances, held out from the SFT splits, in both a video (TI2V) and an interleaved-image setting.

At a glance

Property Value
Tasks 50
Instances per task 1,000
Total instances 50,000 per setting
Archives 50 video + 50 image tar.gz
Total size 10.9 GB (video 9.7 + image 1.2)
Resolution 1024 × 1024, as generated
Video 16 fps, MPEG-4 Part 2, no re-encode

Layout

.
├── VBVR-Pro-RL-Video/
│   ├── G-131_select_next_figure_increasing_size_sequence_data-generator.tar.gz
│   └── …                                                     # 50 archives
├── VBVR-Pro-RL-Image/
│   ├── G-131_select_next_figure_increasing_size_sequence_data-generator.tar.gz
│   └── …                                                     # 50 archives
├── video_jsonl/                                              # 50 files
├── image_jsonl/                                              # 50 files
├── meta_video_rl.json
└── meta_image_rl.json

An archive under VBVR-Pro-RL-Video/ extracts to:

G-131_select_next_figure_increasing_size_sequence_data-generator/        # task
└── select_next_figure_increasing_size_sequence_task/                    # subtask
    ├── select_next_figure_increasing_size_sequence_00005005/            # sample
    │   ├── first_frame.png          # conditioning image  (1024 × 1024)
    │   ├── metadata.json            # task parameters, ground truth, scoring contract
    │   └── video/
    │       ├── prompt.txt           # instruction
    │       ├── ground_truth.mp4     # reference video     (16 fps, MPEG-4 Part 2)
    │       └── final_frame.png      # last frame of the reference video
    ├── select_next_figure_increasing_size_sequence_00005006/            # same files
    ├── …
    └── select_next_figure_increasing_size_sequence_00006004/            # 1,000 samples per task

The matching archive under VBVR-Pro-RL-Image/ has the same task, subtask and sample names, with the reference output as an image sequence instead:

G-131_select_next_figure_increasing_size_sequence_data-generator/
└── select_next_figure_increasing_size_sequence_task/
    ├── select_next_figure_increasing_size_sequence_00005005/
    │   ├── first_frame.png          # conditioning image  (1024 × 1024)
    │   ├── metadata.json            # task parameters, ground truth, scoring contract
    │   └── image/
    │       ├── prompt.txt           # instruction
    │       ├── frame_1.png          # reference output, step 1
    │       └── …                    # up to frame_N.png
    ├── select_next_figure_increasing_size_sequence_00005006/
    ├── …
    └── select_next_figure_increasing_size_sequence_00006004/

Usage

huggingface-cli download Video-Reason/VBVR-Pro-RL --repo-type dataset --local-dir ./rl
mkdir -p rl-data && for f in ./rl/VBVR-Pro-RL-Video/*.tar.gz; do tar xzf "$f" -C rl-data; done

License

VBVR-Pro source code, scripts, configuration files and task-specific scoring software — including everything in this repository — are licensed under the Apache License 2.0. VBVR-Pro data and benchmark materials are separately licensed under CC BY-NC 4.0. Model weights and third-party materials remain subject to their applicable model-card and upstream terms. See LICENSE.md for details.

Citation

@misc{xu2026vbvrproscalableverifiablesuite,
      title={VBVR-Pro: A Scalable and Verifiable Suite for Native Visual Reasoning},
      author={Junxiang Xu and Ruisi Wang and Fanyi Pu and Maijunxian Wang and Ran Ji and Tongxi Zhou and Chenyang Gu and Jing Zuo and Hongcan Xiao and Yimeng Geng and Wanqi Yin and Wei Chen and Oscar Qian and Zhengan Yan and Ziqi Huang and Haiwen Diao and Liang Pan and Bo Li and Xiangyu Fan and Dezhi Luo and Fengyuan Yu and Zehong Zhao and Qingying Gao and Tinghui Zhu and Yilan Zhang and Jingqi Tong and Pinyuan Feng and Zhengze Jiang and Letian Wang and Ziyu Guo and Renrui Zhang and Jieneng Chen and Sonia Joseph and Constantin Venhoff and Saman Motamed and Mengyue Yang and Chandra Sripada and Alan Yuille and Philip Torr and Lvmin Zhang and Vikash Kumar and Daniel Khashabi and Nikolaus Kriegeskorte and Rapha\"{e}l Milli\`{e}re and Vincent C. M\"{u}ller and Anyi Rao and Quan Wang and Ziwei Liu and Dahua Lin and Lei Yang and Hokin Deng and Zhongang Cai},
      year={2026},
      eprint={2608.26105},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2608.26105},
}