We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:


References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computer Vision and Pattern Recognition

Title: Data-Efficient Instance Segmentation with a Single GPU

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
Comments: Technical report
Subjects: Computer Vision and Pattern Recognition (cs.CV)
Cite as: arXiv:2110.00242 [cs.CV]
  (or arXiv:2110.00242v1 [cs.CV] for this version)

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)

Link back to: arXiv, form interface, contact.