YAML Metadata Warning:The task_categories "video-to-3d" is not in the official list: text-classification, token-classification, table-question-answering, question-answering, zero-shot-classification, translation, summarization, feature-extraction, text-generation, fill-mask, sentence-similarity, text-to-speech, text-to-audio, automatic-speech-recognition, audio-to-audio, audio-classification, audio-text-to-text, voice-activity-detection, depth-estimation, image-classification, object-detection, image-segmentation, text-to-image, image-to-text, image-to-image, image-to-video, unconditional-image-generation, video-classification, reinforcement-learning, robotics, tabular-classification, tabular-regression, tabular-to-text, table-to-text, multiple-choice, text-ranking, text-retrieval, time-series-forecasting, text-to-video, image-text-to-text, image-text-to-image, image-text-to-video, visual-question-answering, document-question-answering, zero-shot-image-classification, graph-ml, mask-generation, zero-shot-object-detection, text-to-3d, image-to-3d, image-feature-extraction, video-text-to-text, keypoint-detection, visual-document-retrieval, any-to-any, video-to-video, other

HAT4D: Human-Assisted Training for 4D Dynamic Scene Understanding

MVOIK-4D: Multi-View Object Interaction Knowledge for 4D Physical Reasoning

If you find this dataset useful, please consider citing our paper and following the project page for updates.

💡 Description

MVOIK-4D is a curated release of real-world object-interaction sequences for 4D scene understanding, reconstruction, and evaluation. It contains RGB input frames, multi-view evaluation frames, and memory-mask annotations organized by interaction case.

This dataset is released with the HAT4D project, which studies 4D reasoning and reconstruction for real physical object interactions with temporally consistent dynamic scene understanding.

Project Page: HAT4D
Paper: arXiv:2606.28215
License: CC BY-NC 4.0
Tasks: Image-to-3D, Video-to-3D, 4D dynamic scene understanding, physical reasoning

📁 Structure

The release contains three folder-based splits:

with_gt/: real interaction cases with RGB input frames and multi-view evaluation ground truth.
with_gt_memory/: cases corresponding to the memory setting.
with_gt_memory_mask/: mask annotations aligned with with_gt_memory.

Each split follows the same case-level layout:

<split_name>/
  input/
    <case_id>_<case_name>/
      *.png
  eval_gt/
    <case_id>_<case_name>/
      eval_0/
        *.png
      eval_1/
        *.png
      eval_2/
        *.png
      eval_3/
        *.png

Case folders are named with a four-digit ID followed by an interaction name:

0000_cut_apple_1
0027_pile_bricks_5
0111_box_cover_threads_1

🚀 Usage

After downloading the dataset, load the PNG frames directly from the folder structure:

input/<case>/: input RGB frame sequence for each interaction case.
eval_gt/<case>/eval_0..eval_3/: evaluation views or ground-truth frame sequences.
with_gt_memory_mask/: memory-mask annotations using the same case naming and view structure as with_gt_memory/.

For more details, please refer to the HAT4D project page and paper: Project Page | arXiv

✏️ Citation

If you find this dataset useful in your research, please cite the HAT4D paper:

@inproceedings{Li2026hat4d,
  title     = {HAT-4D: Lifting Monocular Video for 4D Multi-Object Interactions via Human-Agent Collaboration},
  author    = {Li, Jiaxin and Author, Second and Author, Third},
  booktitle = {Computer Vision -- ECCV 2026},
  year      = {2026},
  note      = {Accepted, to appear}
}

Please refer to the arXiv page for the final citation information.

Downloads last month: 75

Total file size:

5.71 kB

Paper for Lijiaxin0111/Open_MVOIK-4D

HAT-4D: Lifting Monocular Video for 4D Multi-Object Interactions via Human-Agent Collaboration

Paper • 2606.28215 • Published 19 days ago