Datasets:
metadata
license: apache-2.0
task_categories:
- feature-extraction
tags:
- disentanglement
- sequential_disentanglement
pretty_name: SMD Sprites
size_categories:
- 10K<n<100K
dataset_info:
features:
- name: id
dtype: int32
- name: x
sequence: image
- name: movement
dtype:
class_label:
names:
'0': front walk
'1': left walk
'2': right walk
'3': front hand raise
'4': left hand raise
'5': right hand raise
'6': front wave
'7': left wave
'8': right wave
- name: body
dtype:
class_label:
names:
'0': light
'1': neutral
'2': dark-gray
'3': gray
'4': brown
'5': black
- name: bottom
dtype:
class_label:
names:
'0': white
'1': yellow
'2': red
'3': gray
'4': shade-green
'5': green
- name: top
dtype:
class_label:
names:
'0': red
'1': blue
'2': white
'3': gray
'4': brown
'5': white-tie
- name: hair
dtype:
class_label:
names:
'0': green
'1': blue
'2': orange
'3': white
'4': red
'5': purple
splits:
- name: train
num_bytes: 110472147.432
num_examples: 8164
- name: val
num_bytes: 24037860.5
num_examples: 1750
- name: test
num_bytes: 24098037.5
num_examples: 1750
download_size: 140877104
dataset_size: 158608045.43199998
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: val
path: data/val-*
- split: test
path: data/test-*
SMD Sprites Dataset Attribution
This benchmark includes a modified variant of the Sprites dataset, adapted to support sequential multi-factor disentanglement.
Original repository:
https://github.com/YingzhenLi/SpritesReference paper:
Y. Li, J. Yu, C. Zhang, C. Gan.
Disentangled Sequential Autoencoder, ICML 2018.
https://arxiv.org/abs/1803.02991
⚠ Note: The Sprites dataset does not have an explicit open-source license. 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.