license: cc-by-nc-4.0
pretty_name: VSCD
task_categories:

• image-segmentation
• video-classification
  size_categories:
• 100GB<n<1TB

VSCD: Video-based Scene Change Detection in Unaligned Scenes

Dataset Description

VSCD is a benchmark for Video-based Scene Change Detection. Given two RGB videos of the same indoor environment captured at different scene states, the task is to predict a pixel-wise change mask aligned with the query video.

Each environment contains five scene videos, scene0.mp4 to scene4.mp4. A VSCD sample is defined by a directed scene pair:

scene0 -> scene1
scene1 -> scene2
scene2 -> scene3
scene3 -> scene4
scene4 -> scene0

For each directed pair sceneA -> sceneB, sceneA is used as the reference video, sceneB is used as the query video, and change_mask_sceneA_to_sceneB.mp4 provides the query-aligned binary change-mask video.

The benchmark is designed for unaligned video pairs with unconstrained camera motion, strong viewpoint mismatch, occlusions, and multiple object-level changes. Object appearances, disappearances, and relocations are treated as changes. Appearance-only variations such as illumination, shadows, and reflections are not considered semantic changes.

Dataset Structure

The dataset is organized into train and test splits:

vscd/
├── train/
│   ├── train_<EnvironmentName>/
│   │   ├── scene0.mp4
│   │   ├── scene1.mp4
│   │   ├── scene2.mp4
│   │   ├── scene3.mp4
│   │   ├── scene4.mp4
│   │   ├── change_mask_scene0_to_scene1.mp4
│   │   ├── change_mask_scene1_to_scene2.mp4
│   │   ├── change_mask_scene2_to_scene3.mp4
│   │   ├── change_mask_scene3_to_scene4.mp4
│   │   └── change_mask_scene4_to_scene0.mp4
│   └── ...
└── test/
    ├── test_<EnvironmentName>/
    │   ├── scene0.mp4
    │   ├── scene1.mp4
    │   ├── scene2.mp4
    │   ├── scene3.mp4
    │   ├── scene4.mp4
    │   ├── change_mask_scene0_to_scene1.mp4
    │   ├── change_mask_scene1_to_scene2.mp4
    │   ├── change_mask_scene2_to_scene3.mp4
    │   ├── change_mask_scene3_to_scene4.mp4
    │   └── change_mask_scene4_to_scene0.mp4
    └── ...

For example, an environment folder may look like:

test/test_BedroomScene18/
├── scene0.mp4
├── scene1.mp4
├── scene2.mp4
├── scene3.mp4
├── scene4.mp4
├── change_mask_scene0_to_scene1.mp4
├── change_mask_scene1_to_scene2.mp4
├── change_mask_scene2_to_scene3.mp4
├── change_mask_scene3_to_scene4.mp4
└── change_mask_scene4_to_scene0.mp4

Splits

• train: training split
• test: test split

Each split contains multiple environment folders. Each environment folder defines five directed video-pair samples.

Pair Definition

The dataset uses the following fixed directed pairs:

For a pair sceneA -> sceneB:

• sceneA.mp4: reference video
• sceneB.mp4: query video
• change_mask_sceneA_to_sceneB.mp4: binary change mask video aligned with sceneB.mp4

Download

To download the full dataset:

hf download jiae1234/vscd --repo-type dataset --local-dir ./VSCD

To download a specific file:

hf download jiae1234/vscd test/test_BedroomScene18/scene0.mp4 --repo-type dataset --local-dir ./VSCD_sample

License

This dataset is released under the Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0).

The dataset may be used, shared, and adapted for non-commercial research purposes with proper attribution.

Asset Notice

This dataset contains only rendered video files and rendered change-mask annotations generated from simulation environments. It does not include or redistribute the original 3D assets, textures, meshes, Unreal Engine project files, simulator source assets, or asset-pack files used to generate the videos.

Some rendered environments were generated using AI2-THOR and Unreal Engine/Fab/Marketplace assets. Users of this dataset are granted access only to the rendered video data and annotations released in this repository, not to the underlying third-party assets.

Citation

If you use this dataset, please cite:

@inproceedings{yoon2026vscd,
  title={VSCD: Video-based Scene Change Detection in Unaligned Scenes},
  author={Yoon, Jiae and Kim, Ue-Hwan},
  booktitle={Proceedings of the 43rd International Conference on Machine Learning},
  year={2026}
}