---
license: apache-2.0
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: val
    path: data/val-*
  - split: test
    path: data/test-*
dataset_info:
  features:
  - name: id
    dtype: int32
  - name: x
    sequence: image
  - name: floor_hue
    dtype:
      class_label:
        names:
          '0': '0.0'
          '1': '0.1'
          '2': '0.2'
          '3': '0.3'
          '4': '0.4'
          '5': '0.5'
          '6': '0.6'
          '7': '0.7'
          '8': '0.8'
          '9': '0.9'
  - name: wall_hue
    dtype:
      class_label:
        names:
          '0': '0.0'
          '1': '0.1'
          '2': '0.2'
          '3': '0.3'
          '4': '0.4'
          '5': '0.5'
          '6': '0.6'
          '7': '0.7'
          '8': '0.8'
          '9': '0.9'
  - name: object_hue
    dtype:
      class_label:
        names:
          '0': '0.0'
          '1': '0.1'
          '2': '0.2'
          '3': '0.3'
          '4': '0.4'
          '5': '0.5'
          '6': '0.6'
          '7': '0.7'
          '8': '0.8'
          '9': '0.9'
  - name: shape
    dtype:
      class_label:
        names:
          '0': cube
          '1': cylinder
          '2': sphere
          '3': capsule
  - name: scale_dynamic
    dtype:
      class_label:
        names:
          '0': increasing_1x
          '1': decreasing_1x
          '2': increasing_2x
          '3': decreasing_2x
          '4': alternating_big_small
          '5': alternating_small_big
  - name: orientation_dynamic
    dtype:
      class_label:
        names:
          '0': counterclockwise
          '1': static
          '2': clockwise
  splits:
  - name: train
    num_bytes: 1719272462.0
    num_examples: 50400
  - name: val
    num_bytes: 386294024.0
    num_examples: 10800
  - name: test
    num_bytes: 385308154.0
    num_examples: 10800
  download_size: 1992990038
  dataset_size: 2490874640.0
---


## MSD Shapes3D Dataset Attribution

The Multi-factor Sequential Disentanglement benchmark includes a **modified variant of the 3D Shapes dataset**, adapted to support sequential multi-factor disentanglement.

In this setup, the **floor hue**, **wall hue**, **object hue**, and **shape** are static, while the **scale** and **orientation** change over time.


- Original repository:  
  https://github.com/deepmind/3d-shapes

```
@misc{3dshapes18,
  title={3D Shapes Dataset},
  author={Burgess, Chris and Kim, Hyunjik},
  howpublished={https://github.com/deepmind/3dshapes-dataset/},
  year={2018}
}
```

⚠ **Note:** The 3D Shapes dataset is released under the Apache License 2.0. We redistribute it here solely for non-commercial research purposes, following the original publication. Please cite the above paper when using this dataset in your work.