[Video download] Due to Policy constraints we are not able to directly provide and host videos. However, we thank Tarun Kalluri who is also working on the dataset and willing to share his pre-processed videos that match baseline performances: Google Drive Link. You can use our provided video2frames.py script to split videos into frames.
UVO v1.0 contains:
Sparse Annotation (UVO-Sparse 1.0):
Compare to v0.5, we add the downsampled version of dense annotations of v1.0. The downsampled dense split contains two classes: "object" for COCO categories and "other" for non-COCO categories, which can be used to test generalization ability.
5641 videos for training (5138 sparsely annotated at 1fps, 503 from video-dense)
2708 videos for validation (2452 sparsely annotated at 1fps, 256 from video-dense)
2879 (2623 sparsely annotated, 256 from video-dense) videos for test, used for Challenge
Interpolation for 30fps is provided (for sparsely annotated splits)
Dense Annotation (UVO-Dense): 503 videos for training and 256 videos for validation, annotated at 30fps
258 videos for test, used for Challenge (video meta data to be released)
COCO taxonomy labels for all objects (marked as "other" if not in COCO categories)
For instructions and baselines on the dataset, we refer to README.md in the shared folder.
Test server at evalai:
Frame: https://eval.ai/web/challenges/challenge-page/1835/overview
Video: https://eval.ai/web/challenges/challenge-page/1840/overview
[Update Sep 10th] We received common requests for deleted videos and pre-processed videos to match the annotations. Due to Policy constraints we are not able to directly provide and host videos. However, we thank Tarun Kalluri who is also working on the dataset and willing to share his pre-processed videos that match baseline performances: Google Drive Link. You can use our provided video2frames.py script to split videos into frames. This will be a transient solution while we look for best way to host dataset videos. Thanks for the support from the community!
UVO v0.5 is a subset of full UVO Dataset. The goal is give researchers a more complete taste of the dataset to start working on the dataset.
UVO v0.5 contains:
Sparse Annotation (UVO-Sparse): 5138 videos for training and 2452 videos for validation, annotated at 1fps
Interpolation for 30fps is provided
Contains 100% sparse annotation planned to release
2623 videos for test, used for 2021 Challenge
Dense Annotation (UVO-Dense): 262 videos for training and 129 videos for validation, annotated at 30fps
142 videos for test, used for 2021 Challenge
The other 50% will be released soon
For instructions and baselines on the dataset, we refer to README.md in the shared folder.
Full training and validation videos and annotations will be released soon! Stay tuned!
Testing videos and test server are out on EvalAI, Track 1 Server, Track 2 Server
UVO PlaySet is a subset of UVO training videos. The goal is to get researchers familiar with the dataset and prototype some models on the dataset.
UVO PlaySet contains:
Sparse Annotation: 4250 videos annotated at 1fps
Dense Annotation: 262 videos annotated at 30fps
For instructions on the dataset, we refer to README.md in the shared folder
Full training and validation set will be released soon! Stay tuned!