References & Citations
Computer Science > Computer Vision and Pattern Recognition
Title: Data-Efficient Instance Segmentation with a Single GPU
(Submitted on 1 Oct 2021 (v1), revised 8 Oct 2021 (this version, v2), latest version 5 Nov 2022 (v5))
Abstract: Not everyone is wealthy enough to have hundreds of GPUs or TPUs. Therefore, we've got to find a way out. In this paper, we introduce a data-efficient instance segmentation method we used in the 2021 VIPriors Instance Segmentation Challenge. Our solution is a modified version of Swin Transformer, based on the mmdetection which is a powerful toolbox. To solve the problem of lack of data, we utilize data augmentation including random flip and multiscale training to train our model. During inference, multiscale fusion is used to boost the performance. We only use a single GPU during the whole training and testing stages. In the end, our team named THU_IVG_2018 achieved the result of 0.366 for AP@0.50:0.95 on the test set, which is competitive with other top-ranking methods while only one GPU is used. Besides, our method achieved the AP@0.50:0.95 (medium) of 0.592, which ranks second among all contestants. In the end, our team ranked third among all the contestants, as announced by the organizers.
Submission history
From: Wanhua Li [view email][v1] Fri, 1 Oct 2021 07:36:20 GMT (135kb,D)
[v2] Fri, 8 Oct 2021 16:31:52 GMT (135kb,D)
[v3] Sun, 14 Nov 2021 01:58:36 GMT (135kb,D)
[v4] Wed, 2 Nov 2022 21:29:07 GMT (0kb,I)
[v5] Sat, 5 Nov 2022 14:29:30 GMT (0kb,I)
Link back to: arXiv, form interface, contact.