%0 Conference Proceedings
%T Learning Multi-Object Tracking and Segmentation from Automatic Annotations
%A Lorenzo Porzi
%A Markus Hofinger
%A Idoia Ruiz
%A Joan Serrat
%A Samuel Rota Bulo
%A Peter Kontschieder
%B 33rd IEEE Conference on Computer Vision and Pattern Recognition
%D 2020
%F Lorenzo Porzi2020
%O ADAS; 600.124; 600.118
%O exported from refbase (http://refbase.cvc.uab.es/show.php?record=3402), last updated on Tue, 16 Nov 2021 13:32:28 +0100
%X In this work we contribute a novel pipeline to automatically generate training data, and to improve over state-of-the-art multi-object tracking and segmentation (MOTS) methods. Our proposed track mining algorithm turns raw street-level videos into high-fidelity MOTS training data, is scalable and overcomes the need of expensive and time-consuming manual annotation approaches. We leverage state-of-the-art instance segmentation results in combination with optical flow predictions, also trained on automatically harvested training data. Our second major contribution is MOTSNet – a deep learning, tracking-by-detection architecture for MOTS – deploying a novel mask-pooling layer for improved object association over time. Training MOTSNet with our automatically extracted data leads to significantly improved sMOTSA scores on the novel KITTI MOTS dataset (+1.9%/+7.5% on cars/pedestrians), and MOTSNet improves by +4.1% over previously best methods on the MOTSChallenge dataset. Our most impressive finding is that we can improve over previous best-performing works, even in complete absence of manually annotated MOTS training data.
%U https://ieeexplore.ieee.org/abstract/document/9157138
%U http://refbase.cvc.uab.es/files/PHR2020.pdf
%U http://dx.doi.org/10.1109/cvpr42600.2020.00688
%P 6845-6854