Datasets:

aRy4n
/

eXtreme-Deformable

license: cc-by-4.0
task_categories:
  - image-text-to-image
  - depth-estimation
  - image-segmentation
  - video-classification
  - visual-question-answering
  - image-feature-extraction
  - image-to-3d
language:
  - en
tags:
  - Burst-Reconstruction
  - Deformable-Scenes
  - High-Speed
  - Non-lambertian
  - Rendering-Physics
pretty_name: XD
size_categories:
  - n<1K

eXtreme-Deformable (XD) Evaluation Dataset from gQIR: Generative Quanta Image Reconstruction

Aryan Garg¹, Sizhuo Ma², Mohit Gupta¹

¹ University of Wisconsin-Madison
² Snap, Inc

Interesting deformable physics scenes - explosions, shattering objects, bullets, fluids etc scraped from the internet. All the videos have been manually filtered for interesting content (subjective binary decision) and then processed: removed audio --> clipped --> resized. Each video is atleast 10 seconds or > 200-frames. See SPAD-Simulation-Pipeline for streaming the input to train/test models.

Credits to original creators with majority scrape results:

NOTE: So far only paper + supplementary GT videos are added. Full dataset will be released gradually with outputs and more testing.

Tentative: June 1, 2026 for full XD-dataset release.

Please cite our dataset/work if you find it useful. Thanks! :)

@InProceedings{garg_2026_gqir,
    author    = {Garg, Aryan and Ma, Sizhuo and  Gupta, Mohit},
    title     = {gQIR: Generative Quanta Image Reconstruction},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2026},
}

gQIR's code & weights are available at: Github & HF respectively.