Temporal RoI Align for Video Object Recognition

8 Sep 2021  ·  Tao Gong, Kai Chen, Xinjiang Wang, Qi Chu, Feng Zhu, Dahua Lin, Nenghai Yu, Huamin Feng ·

Video object detection is challenging in the presence of appearance deterioration in certain video frames. Therefore, it is a natural choice to aggregate temporal information from other frames of the same video into the current frame. However, RoI Align, as one of the most core procedures of video detectors, still remains extracting features from a single-frame feature map for proposals, making the extracted RoI features lack temporal information from videos. In this work, considering the features of the same object instance are highly similar among frames in a video, a novel Temporal RoI Align operator is proposed to extract features from other frames feature maps for current frame proposals by utilizing feature similarity. The proposed Temporal RoI Align operator can extract temporal information from the entire video for proposals. We integrate it into single-frame video detectors and other state-of-the-art video detectors, and conduct quantitative experiments to demonstrate that the proposed Temporal RoI Align operator can consistently and significantly boost the performance. Besides, the proposed Temporal RoI Align can also be applied into video instance segmentation. Codes are available at https://github.com/open-mmlab/mmtracking

PDF Abstract
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Video Object Detection EPIC KITCHENS-seen splits Temporal ROI Align mAP 42.2 # 1
Video Object Detection EPIC KITCHENS-unseen splits Temporal ROI Align mAP 39.6 # 1
Video Object Detection ImageNet VID Temporal ROI Align (ResNeXt101) MAP 84.3 # 12
Video Instance Segmentation YouTube-VIS Temporal ROI Align mask AP 38 # 1