A Simple Framework for Contrastive Learning of Visual Representations

Ting Chen, Simon Kornblith, Mohammad Norouzi, Geoffrey Hinton

arXiv · PDF · Open in the Atlas

Code that ran

We lifted 137 functions out of this paper's own repositories and ran 115 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.

Showing the first 60. An agent gets every one from get_harvested_code_for_paper("2002.05709").

Repositories linked to this paper

• google-research/simclr — marked official in Papers with Code; named in the paper
• 63days/SimCLR
• AidenDurrant/MoCo-Pytorch
• AndrewAtanov/simclr-pytorch
• BalajiAI/SimCLR
• Daniel-H-99/SimCLR
• HobbitLong/SupContrast
• JoonHyung-Park/SimCLR
• Liut2016/ecg-supcontrast
• Mind23-2/MindCode-76
• SaeedShurrab/SimSiam-pytorch
• Westlake-AI/openmixup
• XG293/SupConLoss
• acholston/SimCLR
• ae-foster/pytorch-simclr
• alexk1704/scclv2
• anonymousmlsubmission/datadistributiontransfer
• applecrumble123/SimCLR_pytorch
• asd08573064/SimCLR
• beresandras/contrastive-classification-keras
• bjader/QSSL
• brainimageanalysis/marmosetgeneatlas_adult
• caesarea38/doclangid
• chingyaoc/DCL
• cjrd/self-supervised-pretraining
• clabrugere/pytorch-scarf
• cyrilvallez/image-manipulation-detection
• danielzgsilva/MOT
• davidczy/supcon_gamma
• delyan-boychev/grad-cache-con-learning

More are linked than shown here.

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

This paper presents SimCLR: a simple framework for contrastive learning of visual representations. We simplify recently proposed contrastive selfsupervised learning algorithms without requiring specialized architectures or a memory bank. In order to understand what enables the contrastive prediction tasks to learn useful representations, we systematically study the major components of our framework. We show that (1) composition of data augmentations plays a critical role in defining effective predictive tasks, (2) introducing a learnable nonlinear transformation between the representation and the contrastive loss substantially improves the quality of the learned representations, and (3) contrastive learning benefits from larger batch sizes and more training steps compared to supervised learning. By combining these findings, we are able to considerably outperform previous methods for self-supervised and semi-supervised learning on ImageNet. A linear classifier trained on self-supervised representations learned by Sim-CLR achieves 76.5% top-1 accuracy, which is a 7% relative improvement over previous state-ofthe-art, matching the performance of a supervised ResNet-50. When fine-tuned on only 1% of the labels, we achieve 85.8% top-5 accuracy, outperforming AlexNet with 100× fewer labels. 1

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2002.05709")
get_code_for_paper("2002.05709")
have("2002.05709")

Connect an agent — have() is free.