OVIS (short for Occluded Video Instance Segmentation) is a new large scale benchmark dataset for video instance segmentation task. It is designed with the philosophy of perceiving object occlusions in videos, which could reveal the complexity and the diversity of real-world scenes.
Occluded Video Instance Segmentation
Jiyang Qi1,2, Yan Gao2, Yao Hu2, Xinggang Wang1, Xiaoyu Liu2, Xiang Bai1, Serge Belongie3, Alan Yuille4, Philip H.S. Torr5, Song Bai2,5
1Huazhong University of Science and Technology 2Alibaba Group 3Cornell University
4Johns Hopkins University 5University of Oxford
Can our video understanding systems perceive objects when a heavy occlusion exists in a scene?
To answer this question, we collect a large scale dataset called OVIS for occluded video instance segmentation, that is, to simultaneously detect, segment, and track instances in occluded scenes. OVIS consists of 296k high-quality instance masks from 25 semantic categories, where object occlusions usually occur. While our human vision systems can understand those occluded instances by contextual reasoning and association, our experiments suggest that current video understanding systems are not satisfying. On the OVIS dataset, the highest AP achieved by state-of-the-art algorithms is only 14.4, which reveals that we are still at a nascent stage for understanding objects, instances, and videos in a real-world scenario. Moreover, to complement missing object cues caused by occlusion, we propose a plug-and-play module called temporal feature calibration. Built upon MaskTrack R-CNN and SipMask, we report an AP of 15.2 and 15.0 respectively.
OVIS Consists of:
- 296k high-quality instance masks
- 25 commonly seen semantic categories
- 901 videos with severe object occlusions
- 5,223 unique instances
Given a video, all the objects belonging to the pre-defined category set are exhaustively annotated. All the videos are annotated per 5 frames.
- Severe occlusions. The most distinctive property of our OVIS dataset is that a large portion of objects is under various types of severe occlusions caused by different factors.
- Long videos. The average video duration and the average instance duration of OVIS are 12.77s and 10.05s respectively.
- Crowded scenes. On average, there are 5.80 instances per video and 4.72 objects per frame.
The 25 semantic categories in OVIS are Person, Bird, Cat, Dog, Horse, Sheep, Cow, Elephant, Bear, Zebra, Giraffe, Poultry, Giant panda, Lizard, Parrot, Monkey, Rabbit, Tiger, Fish, Turtle, Bicycle, Motorcycle, Airplane, Boat, and Vehicle.
For a detailed description of OVIS, please refer to our paper.
We provide the frames and annotations.
- Frames: JPG format. The total size is 12.7GB.
- Annotations: JSON format.
The annotations are COCO-style, just like Youtube-VIS. So it’s nearly cost-free to adapt your Youtube-VIS code for OVIS. (Please refer to this repo for the code of loading annotations.)
The code and models of the baseline method are released on [comming soon].
The evaluation metric is the same as Youtube-VIS’s, so you can use the evalution code provided by them [link]
For questions and suggestions, please contact Jiyang Qi (jiyangqi at hust dot edu dot cn).