タオ

  • 説明:

TAO データセットは、2,907 の高解像度ビデオと 833 のオブジェクト カテゴリで構成される大規模なビデオ オブジェクト検出データセットです。このデータセットを保存するには、少なくとも 300 GB の空き容量が必要であることに注意してください。

  • 追加のドキュメント:コードを使用したペーパーの探索

  • ホームページ: https://taodataset.org/

  • ソースコード: tfds.video.tao.Tao

  • バージョン:

    • 1.0.0 (デフォルト): 初期リリース。
  • ダウンロードサイズ: 113.96 GiB

  • 手動ダウンロードの手順: このデータセットでは、ソース データを手動でdownload_config.manual_dir (デフォルトは~/tensorflow_datasets/downloads/manual/ ) にダウンロードする必要があります。
    一部の TAO ファイル (HVACS および AVA ビデオ) は、MOT へのログインが必要なため、手動でダウンロードする必要があります。 https://motchallenge.net/tao_download.phpの指示に従ってダウンロードしてください。

このデータをダウンロードし、結果の .zip ファイルを ~/tensorflow_datasets/downloads/manual/ に移動します

手動ダウンロードが必要なデータが存在しない場合はスキップされ、手動ダウンロードが不要なデータのみが使用されます。

スプリット
'train' 500
'validation' 988
@article{Dave_2020,
   title={TAO: A Large-Scale Benchmark for Tracking Any Object},
   ISBN={9783030585587},
   ISSN={1611-3349},
   url={http://dx.doi.org/10.1007/978-3-030-58558-7_26},
   DOI={10.1007/978-3-030-58558-7_26},
   journal={Lecture Notes in Computer Science},
   publisher={Springer International Publishing},
   author={Dave, Achal and Khurana, Tarasha and Tokmakov, Pavel and Schmid, Cordelia and Ramanan, Deva},
   year={2020},
   pages={436-454}
}

tao/480_640 (デフォルト設定)

  • 構成の説明: すべての画像はバイリニアで 480 X 640 にサイズ変更されます

  • データセットサイズ: 482.30 GiB

  • 機能構造:

FeaturesDict({
    'metadata': FeaturesDict({
        'dataset': string,
        'height': int32,
        'neg_category_ids': Tensor(shape=(None,), dtype=int32),
        'not_exhaustive_category_ids': Tensor(shape=(None,), dtype=int32),
        'num_frames': int32,
        'video_name': string,
        'width': int32,
    }),
    'tracks': Sequence({
        'bboxes': Sequence(BBoxFeature(shape=(4,), dtype=float32)),
        'category': ClassLabel(shape=(), dtype=int64, num_classes=363),
        'frames': Sequence(int32),
        'is_crowd': bool,
        'scale_category': string,
        'track_id': int32,
    }),
    'video': Video(Image(shape=(480, 640, 3), dtype=uint8)),
})
  • 機能のドキュメント:
特徴クラスDtype説明
特徴辞書
メタデータ特徴辞書
メタデータ/データセットテンソルストリング
メタデータ/高さテンソルint32
メタデータ/neg_category_idsテンソル(なし、) int32
メタデータ/not_exhaustive_category_idsテンソル(なし、) int32
メタデータ/num_framesテンソルint32
メタデータ/ビデオ名テンソルストリング
メタデータ/幅テンソルint32
トラック順序
トラック/bboxシーケンス(BBoxFeature) (なし、4) float32
トラック/カテゴリクラスラベルint64
トラック/フレームシーケンス(テンソル) (なし、) int32
トラック/is_crowdテンソルブール
トラック/scale_categoryテンソルストリング
トラック/track_idテンソルint32
ビデオ動画(画像) (なし、480、640、3) uint8

tao/full_resolution

  • 構成の説明: データセットのフル解像度バージョン。

  • データセットサイズ: 171.24 GiB

  • 機能構造:

FeaturesDict({
    'metadata': FeaturesDict({
        'dataset': string,
        'height': int32,
        'neg_category_ids': Tensor(shape=(None,), dtype=int32),
        'not_exhaustive_category_ids': Tensor(shape=(None,), dtype=int32),
        'num_frames': int32,
        'video_name': string,
        'width': int32,
    }),
    'tracks': Sequence({
        'bboxes': Sequence(BBoxFeature(shape=(4,), dtype=float32)),
        'category': ClassLabel(shape=(), dtype=int64, num_classes=363),
        'frames': Sequence(int32),
        'is_crowd': bool,
        'scale_category': string,
        'track_id': int32,
    }),
    'video': Video(Image(shape=(None, None, 3), dtype=uint8)),
})
  • 機能のドキュメント:
特徴クラスDtype説明
特徴辞書
メタデータ特徴辞書
メタデータ/データセットテンソルストリング
メタデータ/高さテンソルint32
メタデータ/neg_category_idsテンソル(なし、) int32
メタデータ/not_exhaustive_category_idsテンソル(なし、) int32
メタデータ/num_framesテンソルint32
メタデータ/ビデオ名テンソルストリング
メタデータ/幅テンソルint32
トラック順序
トラック/bboxシーケンス(BBoxFeature) (なし、4) float32
トラック/カテゴリクラスラベルint64
トラック/フレームシーケンス(テンソル) (なし、) int32
トラック/is_crowdテンソルブール
トラック/scale_categoryテンソルストリング
トラック/track_idテンソルint32
ビデオ動画(画像) (なし、なし、なし、3) uint8