- 설명 :
3dshapes는 6개의 실제 독립 잠재 요인으로부터 절차적으로 생성된 3D 모양의 데이터세트입니다. 이러한 요소에는 바닥 색상 , 벽 색상 , 객체 색상 , 규모 , 모양 및 방향이 있습니다.
이러한 잠재 요소의 가능한 모든 조합은 정확히 한 번만 존재하며 N = 480000개의 총 이미지를 생성합니다.
잠재 요인 값
- 바닥 색조: [0, 1]에 선형 간격으로 배치된 10개 값
- 벽 색상: [0, 1]에 선형 간격으로 배치된 10개 값
- 객체 색상: [0, 1]에 선형 간격으로 배치된 10개 값
- scale: [0, 1]에 선형 간격으로 배치된 8개 값
- 모양: [0, 1, 2, 3]의 4개 값
- 방향: [-30, 30]의 선형 간격으로 배치된 15개 값
우리는 한 번에 하나의 잠재성을 변경하고(방향에서 시작하여 모양 등) images 배열에 고정된 순서로 이미지를 순차적으로 저장했습니다. 해당 요인의 값은 labels 배열에 동일한 순서로 저장됩니다.
추가 문서 : 코드 가 포함된 논문 탐색
소스 코드 :
tfds.datasets.shapes3d.Builder버전 :
-
2.0.0(기본값): 새로운 분할 API( https://tensorflow.org/datasets/splits )
-
다운로드 크기 :
255.18 MiB데이터세트 크기 :
1.68 GiB자동 캐시 ( 문서 ): 아니요
분할 :
| 나뉘다 | 예 |
|---|---|
'train' | 480,000 |
- 기능 구조 :
FeaturesDict({
'image': Image(shape=(64, 64, 3), dtype=uint8),
'label_floor_hue': ClassLabel(shape=(), dtype=int64, num_classes=10),
'label_object_hue': ClassLabel(shape=(), dtype=int64, num_classes=10),
'label_orientation': ClassLabel(shape=(), dtype=int64, num_classes=15),
'label_scale': ClassLabel(shape=(), dtype=int64, num_classes=8),
'label_shape': ClassLabel(shape=(), dtype=int64, num_classes=4),
'label_wall_hue': ClassLabel(shape=(), dtype=int64, num_classes=10),
'value_floor_hue': float32,
'value_object_hue': float32,
'value_orientation': float32,
'value_scale': float32,
'value_shape': float32,
'value_wall_hue': float32,
})
- 기능 문서 :
| 특징 | 수업 | 모양 | Dtype | 설명 |
|---|---|---|---|---|
| 특징Dict | ||||
| 영상 | 영상 | (64, 64, 3) | uint8 | |
| label_floor_hue | 클래스 라벨 | 정수64 | ||
| label_object_hue | 클래스 라벨 | 정수64 | ||
| label_orientation | 클래스 라벨 | 정수64 | ||
| label_scale | 클래스 라벨 | 정수64 | ||
| label_shape | 클래스 라벨 | 정수64 | ||
| label_wall_hue | 클래스 라벨 | 정수64 | ||
| 값_바닥_색조 | 텐서 | float32 | ||
| value_object_hue | 텐서 | float32 | ||
| 가치지향 | 텐서 | float32 | ||
| 가치 규모 | 텐서 | float32 | ||
| value_shape | 텐서 | float32 | ||
| 값_벽_색조 | 텐서 | float32 |
감독되는 키 (
as_superviseddoc 참조):None그림 ( tfds.show_examples ):

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