CoolFace
Datasetpublic

TalBarami/msd_3dshapes

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}… See the full description on the dataset page: https://huggingface.co/datasets/TalBarami/msd_3dshapes.

sourceHugging Faceapache-2.0updated 1y agoView on Hugging Face
0likes21downloads
README.md122 linesDownload Raw Back to root
1---2license: apache-2.03configs:4- config_name: default5  data_files:6  - split: train7    path: data/train-*8  - split: val9    path: data/val-*10  - split: test11    path: data/test-*12dataset_info:13  features:14  - name: id15    dtype: int3216  - name: x17    sequence: image18  - name: floor_hue19    dtype:20      class_label:21        names:22          '0': '0.0'23          '1': '0.1'24          '2': '0.2'25          '3': '0.3'26          '4': '0.4'27          '5': '0.5'28          '6': '0.6'29          '7': '0.7'30          '8': '0.8'31          '9': '0.9'32  - name: wall_hue33    dtype:34      class_label:35        names:36          '0': '0.0'37          '1': '0.1'38          '2': '0.2'39          '3': '0.3'40          '4': '0.4'41          '5': '0.5'42          '6': '0.6'43          '7': '0.7'44          '8': '0.8'45          '9': '0.9'46  - name: object_hue47    dtype:48      class_label:49        names:50          '0': '0.0'51          '1': '0.1'52          '2': '0.2'53          '3': '0.3'54          '4': '0.4'55          '5': '0.5'56          '6': '0.6'57          '7': '0.7'58          '8': '0.8'59          '9': '0.9'60  - name: shape61    dtype:62      class_label:63        names:64          '0': cube65          '1': cylinder66          '2': sphere67          '3': capsule68  - name: scale_dynamic69    dtype:70      class_label:71        names:72          '0': increasing_1x73          '1': decreasing_1x74          '2': increasing_2x75          '3': decreasing_2x76          '4': alternating_big_small77          '5': alternating_small_big78  - name: orientation_dynamic79    dtype:80      class_label:81        names:82          '0': counterclockwise83          '1': static84          '2': clockwise85  splits:86  - name: train87    num_bytes: 1719272462.088    num_examples: 5040089  - name: val90    num_bytes: 386294024.091    num_examples: 1080092  - name: test93    num_bytes: 385308154.094    num_examples: 1080095  download_size: 199299003896  dataset_size: 2490874640.097---98 99 100## MSD Shapes3D Dataset Attribution101 102The Multi-factor Sequential Disentanglement benchmark includes a **modified variant of the 3D Shapes dataset**, adapted to support sequential multi-factor disentanglement.103 104In this setup, the **floor hue**, **wall hue**, **object hue**, and **shape** are static, while the **scale** and **orientation** change over time.105 106 107- Original repository:  108  https://github.com/deepmind/3d-shapes109 110```111@misc{3dshapes18,112  title={3D Shapes Dataset},113  author={Burgess, Chris and Kim, Hyunjik},114  howpublished={https://github.com/deepmind/3dshapes-dataset/},115  year={2018}116}117```118 119⚠ **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.120 121 122