対称ソリッド,対称ソリッド
コレクションでコンテンツを整理
必要に応じて、コンテンツの保存と分類を行います。
これは姿勢推定データセットであり、複数の方向が視覚的に区別できない対称的な 3D 形状で構成されています。課題は、トレーニング中に 1 つの方向のみが各画像とペアになっている場合に、すべての同等の方向を予測することです (ほとんどの姿勢推定データセットのシナリオと同様)。ほとんどの姿勢推定データセットとは対照的に、同等の向きの完全なセットを評価に使用できます。
合計 8 つの形状があり、それぞれが 3D 回転の全空間にわたって均一にランダムに分散された 50,000 の視点からレンダリングされます。四面体、立方体、二十面体、円錐、円柱の 5 つの形状は特徴がありません。それらのうち、3 つの正多面体 (四面体、立方体、二十面体) には、それぞれ 12 回、24 回、および 60 回の個別の対称性が注釈付けされています。円錐と円柱には、1 度間隔で離散化された連続対称性で注釈が付けられます。これらの対称性は評価のために提供されています。意図した監視は、各画像で 1 回の回転のみです。
残りの 3 つの形状には、特徴的な特徴があります。 1 つの赤い面を持つ四面体、中心から外れた点を持つ円柱、および点で覆われた X を持つ球があります。際立った特徴が目に見えるかどうかにかかわらず、可能な向きのスペースは減少します。これらの形状に相当する回転のセットは提供していません。
各例には以下が含まれます
- 224x224 RGB 画像
データセットを形状でフィルタリングできるようにするための形状インデックス。
インデックスは次のものに対応します。
- 0 = 四面体
- 1 = 立方体
- 2 = 二十面体
- 3 = コーン
- 4 = 円柱
- 5 = マークされた四面体
- 6 = マーク付きの円柱
- 7 = マークされた球
3x3 回転行列として表される、レンダリング プロセスで使用される回転
評価のための、対称性の下での既知の等価回転のセット。
3 つのマークされた形状の場合、これはレンダリングの回転のみです。
スプリット | 例 |
---|
'test' | 40,000 |
'train' | 360,000 |
FeaturesDict({
'image': Image(shape=(224, 224, 3), dtype=uint8),
'label_shape': ClassLabel(shape=(), dtype=int64, num_classes=8),
'rotation': Tensor(shape=(3, 3), dtype=float32),
'rotations_equivalent': Tensor(shape=(None, 3, 3), dtype=float32),
})
特徴 | クラス | 形 | Dtype | 説明 |
---|
| 特徴辞書 | | | |
画像 | 画像 | (224、224、3) | uint8 | |
label_shape | クラスラベル | | int64 | |
回転 | テンソル | (3,3) | float32 | |
同等の回転数 | テンソル | (なし、3、3) | float32 | |

@inproceedings{implicitpdf2021,
title = {Implicit Representation of Probability Distributions on the Rotation
Manifold},
author = {Murphy, Kieran and Esteves, Carlos and Jampani, Varun and
Ramalingam, Srikumar and Makadia, Ameesh}
booktitle = {International Conference on Machine Learning}
year = {2021}
}
特に記載のない限り、このページのコンテンツはクリエイティブ・コモンズの表示 4.0 ライセンスにより使用許諾されます。コードサンプルは Apache 2.0 ライセンスにより使用許諾されます。詳しくは、Google Developers サイトのポリシーをご覧ください。Java は Oracle および関連会社の登録商標です。
最終更新日 2022-11-23 UTC。
[null,null,["最終更新日 2022-11-23 UTC。"],[],[],null,["# symmetric_solids\n\n\u003cbr /\u003e\n\n- **Description**:\n\nThis is a pose estimation dataset, consisting of symmetric 3D shapes where\nmultiple orientations are visually indistinguishable. The challenge is to\npredict all equivalent orientations when only one orientation is paired with\neach image during training (as is the scenario for most pose estimation\ndatasets). In contrast to most pose estimation datasets, the full set of\nequivalent orientations is available for evaluation.\n\nThere are eight shapes total, each rendered from 50,000 viewpoints distributed\nuniformly at random over the full space of 3D rotations. Five of the shapes are\nfeatureless -- tetrahedron, cube, icosahedron, cone, and cylinder. Of those, the\nthree Platonic solids (tetrahedron, cube, icosahedron) are annotated with their\n12-, 24-, and 60-fold discrete symmetries, respectively. The cone and cylinder\nare annotated with their continuous symmetries discretized at 1 degree\nintervals. These symmetries are provided for evaluation; the intended\nsupervision is only a single rotation with each image.\n\nThe remaining three shapes are marked with a distinguishing feature. There is a\ntetrahedron with one red-colored face, a cylinder with an off-center dot, and a\nsphere with an X capped by a dot. Whether or not the distinguishing feature is\nvisible, the space of possible orientations is reduced. We do not provide the\nset of equivalent rotations for these shapes.\n\nEach example contains of\n\n- the 224x224 RGB image\n- a shape index so that the dataset may be filtered by shape. \n\n The indices correspond to:\n\n - 0 = tetrahedron\n - 1 = cube\n - 2 = icosahedron\n - 3 = cone\n - 4 = cylinder\n - 5 = marked tetrahedron\n - 6 = marked cylinder\n - 7 = marked sphere\n- the rotation used in the rendering process, represented as a 3x3 rotation\n matrix\n\n- the set of known equivalent rotations under symmetry, for evaluation.\n\nIn the case of the three marked shapes, this is only the rendering rotation.\n\n- **Additional Documentation** :\n [Explore on Papers With Code\n north_east](https://paperswithcode.com/dataset/symmetric-solids)\n\n- **Homepage** :\n \u003chttps://implicit-pdf.github.io\u003e\n\n- **Source code** :\n [`tfds.datasets.symmetric_solids.Builder`](https://github.com/tensorflow/datasets/tree/master/tensorflow_datasets/datasets/symmetric_solids/symmetric_solids_dataset_builder.py)\n\n- **Versions**:\n\n - **`1.0.0`** (default): Initial release.\n- **Download size** : `3.10 GiB`\n\n- **Dataset size** : `3.94 GiB`\n\n- **Auto-cached**\n ([documentation](https://www.tensorflow.org/datasets/performances#auto-caching)):\n No\n\n- **Splits**:\n\n| Split | Examples |\n|-----------|----------|\n| `'test'` | 40,000 |\n| `'train'` | 360,000 |\n\n- **Feature structure**:\n\n FeaturesDict({\n 'image': Image(shape=(224, 224, 3), dtype=uint8),\n 'label_shape': ClassLabel(shape=(), dtype=int64, num_classes=8),\n 'rotation': Tensor(shape=(3, 3), dtype=float32),\n 'rotations_equivalent': Tensor(shape=(None, 3, 3), dtype=float32),\n })\n\n- **Feature documentation**:\n\n| Feature | Class | Shape | Dtype | Description |\n|----------------------|--------------|---------------|---------|-------------|\n| | FeaturesDict | | | |\n| image | Image | (224, 224, 3) | uint8 | |\n| label_shape | ClassLabel | | int64 | |\n| rotation | Tensor | (3, 3) | float32 | |\n| rotations_equivalent | Tensor | (None, 3, 3) | float32 | |\n\n- **Supervised keys** (See\n [`as_supervised` doc](https://www.tensorflow.org/datasets/api_docs/python/tfds/load#args)):\n `('image', 'rotation')`\n\n- **Figure**\n ([tfds.show_examples](https://www.tensorflow.org/datasets/api_docs/python/tfds/visualization/show_examples)):\n\n- **Examples** ([tfds.as_dataframe](https://www.tensorflow.org/datasets/api_docs/python/tfds/as_dataframe)):\n\nDisplay examples... \n\n- **Citation**:\n\n @inproceedings{implicitpdf2021,\n title = {Implicit Representation of Probability Distributions on the Rotation\n Manifold},\n author = {Murphy, Kieran and Esteves, Carlos and Jampani, Varun and\n Ramalingam, Srikumar and Makadia, Ameesh}\n booktitle = {International Conference on Machine Learning}\n year = {2021}\n }"]]