Prototypical Cross-Attention Networks for Multiple Object Tracking and Segmentation
Open access
Date
2021Type
- Conference Paper
ETH Bibliography
yes
Altmetrics
Abstract
Multiple object tracking and segmentation requires detecting, tracking, and segmenting objects belonging to a set of given classes. Most approaches only exploit the temporal dimension to address the association problem, while relying on single frame predictions for the segmentation mask itself. We propose Prototypical Cross-Attention Network (PCAN), capable of leveraging rich spatio-temporal information for online multiple object tracking and segmentation. PCAN first distills a space-time memory into a set of prototypes and then employs cross-attention to retrieve rich information from the past frames. To segment each object, PCAN adopts a prototypical appearance module to learn a set of contrastive foreground and background prototypes, which are then propagated over time. Extensive experiments demonstrate that PCAN outperforms current video instance tracking and segmentation competition winners on both Youtube-VIS and BDD100K datasets, and shows efficacy to both one-stage and two-stage segmentation frameworks. Code and video resources are available at http://vis.xyz/pub/pcan. Show more
Permanent link
https://doi.org/10.3929/ethz-b-000517118Publication status
publishedExternal links
Editor
Book title
Advances in Neural Information Processing Systems 34Pages / Article No.
Publisher
CurranEvent
Organisational unit
09688 - Yu, Fisher / Yu, Fisher
Notes
Conference lecture held at the poster session on December 8, 2021More
Show all metadata
ETH Bibliography
yes
Altmetrics