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

Download:

Current browse context:

cs.CV

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(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: DisCo: Remedy Self-supervised Learning on Lightweight Models with Distilled Contrastive Learning

Abstract: While self-supervised representation learning (SSL) has received widespread attention from the community, recent research argue that its performance will suffer a cliff fall when the model size decreases. The current method mainly relies on contrastive learning to train the network and in this work, we propose a simple yet effective Distilled Contrastive Learning (DisCo) to ease the issue by a large margin. Specifically, we find the final embedding obtained by the mainstream SSL methods contains the most fruitful information, and propose to distill the final embedding to maximally transmit a teacher's knowledge to a lightweight model by constraining the last embedding of the student to be consistent with that of the teacher. In addition, in the experiment, we find that there exists a phenomenon termed Distilling BottleNeck and present to enlarge the embedding dimension to alleviate this problem. Our method does not introduce any extra parameter to lightweight models during deployment. Experimental results demonstrate that our method achieves the state-of-the-art on all lightweight models. Particularly, when ResNet-101/ResNet-50 is used as teacher to teach EfficientNet-B0, the linear result of EfficientNet-B0 on ImageNet is very close to ResNet-101/ResNet-50, but the number of parameters of EfficientNet-B0 is only 9.4\%/16.3\% of ResNet-101/ResNet-50. Code is available at this https URL com/Yuting-Gao/DisCo-pytorch.
Comments: ECCV 2022
Subjects: Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI)
Cite as: arXiv:2104.09124 [cs.CV]
  (or arXiv:2104.09124v7 [cs.CV] for this version)

Submission history

From: Jia-Xin Zhuang [view email]
[v1] Mon, 19 Apr 2021 08:22:52 GMT (732kb,D)
[v2] Wed, 14 Jul 2021 11:29:35 GMT (732kb,D)
[v3] Mon, 9 Aug 2021 12:28:33 GMT (733kb,D)
[v4] Fri, 18 Feb 2022 06:58:38 GMT (4391kb,D)
[v5] Fri, 27 May 2022 17:08:07 GMT (4391kb,D)
[v6] Mon, 30 May 2022 01:55:54 GMT (4391kb,D)
[v7] Mon, 4 Jul 2022 14:03:59 GMT (4394kb,D)

Link back to: arXiv, form interface, contact.