[2011.09157] Dense Contrastive Learning for Self-Supervised Visual Pre-Trainingopen searchopen navigation menucontact arXivsubscribe to arXiv mailings

To date, most existing self-supervised learning methods are designed and optimized for image classification. These pre-trained models can be sub-optimal for dense prediction tasks due to the discrepancy between image-level prediction and pixel-level prediction. To fill this gap, we aim to design an effective, dense self-supervised learning method that directly works at the level of pixels (or local features) by taking into account the correspondence between local features. We present dense contrastive learning, which implements self-supervised learning by optimizing a pairwise contrastive (dis)similarity loss at the pixel level between two views of input images. Compared to the baseline method MoCo-v2, our method introduces negligible computation overhead (only <1% slower), but demonstrates consistently superior performance when transferring to downstream dense prediction tasks including object detection, semantic segmentation and instance segmentation; and outperforms the state-of-the

1 mentions: @jbohnslav
Date: 2020/11/19 17:21

Referring Tweets

@jbohnslav Dense contrastive learning for self-supervised visual pre-training arxiv: t.co/U6HFy52wwm code: t.co/yfBRiHR788 Self-sup for dense tasks. Uses view-based correspondence to define positive examples. Extends NCELoss to dense outputs (1/2) t.co/COMBk8cXyQ

Related Entries

Read more [2010.09709] Self-supervised Co-training for Video Representation Learningopen searchopen navigation...
0 users, 2 mentions 2020/10/20 11:21
Read more [2010.13938] Neural Unsigned Distance Fields for Implicit Function Learningopen searchopen navigatio...
0 users, 1 mentions 2020/10/28 18:52
Read more [2011.01215v1] Continuous and Diverse Image-to-Image Translation via Signed Attribute Vectorsopen se...
0 users, 1 mentions 2020/11/03 14:21
Read more [2011.02697] Center-wise Local Image Mixture For Contrastive Representation Learningopen searchopen ...
0 users, 1 mentions 2020/11/06 15:51
Read more [2011.11261] Hierarchically Decoupled Spatial-Temporal Contrast for Self-supervised Video Representa...
0 users, 1 mentions 2020/11/24 15:51