AI Paper Atlas

Research map / Label-efficient and robust learning / Semi-supervised learning

CoMatch: Semi-Supervised Learning With Contrastive Graph Regularization

Junnan Li, Caiming Xiong, Steven C.H. Hoi

ICCV 2021 · Label-efficient and robust learning › Semi-supervised learning · approach: Confidence-based pseudo-label selection · 313 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use CoMatch instead of FixMatch w. DAFixMatch w. DA · CIFAR-10 (40 labels)accuracy (%): 86.98 -> 93.09Improved
use CoMatch instead of FixMatch w. DAFixMatch w. DA · STL-10 (1000 labels)accuracy (%): 66.53 -> 79.80Improved
use CoMatch instead of FixMatch w. DAFixMatch w. DA · ImageNet (1% labels, no pre-training)top-1 accuracy (%): 53.4 -> 66.0Improved
use CoMatch instead of FixMatch w. DAFixMatch w. DA · ImageNet (10% labels, no pre-training)top-1 accuracy (%): 70.8 -> 73.6Improved
add distribution alignment to FixMatchFixMatch · CIFAR-10 (40 labels)accuracy (%): 86.12 -> 86.98No clear effect
add distribution alignment to FixMatchFixMatch · STL-10 (1000 labels)accuracy (%): 65.38 -> 66.53Improved
add MoCov2 pre-training to CoMatchCoMatch · ImageNet (1% labels)top-1 accuracy (%): 66.0 -> 67.1Improved
add graph-based pseudo-label connections (T<1) instead of self-loops only (T=1) (ablation)CoMatch · ImageNet (1% labels, 100 epochs)top-1 accuracy (%): decreases the performance by 2.8%Improved
add memory smoothing (alpha=0.9) instead of EMA prediction only (alpha=1) (ablation)CoMatch · ImageNet (1% labels, 100 epochs)top-1 accuracy (%): accuracy decreases by 2.1%Improved
use alpha below 0.9 (stronger smoothing) (ablation)CoMatch · ImageNet (1% labels, 100 epochs)top-1 accuracy (%)Hurt
increase memory bank and queue size K from 10k to 30k (ablation)CoMatch · ImageNet (1% labels, 100 epochs)top-1 accuracy (%)Improved
use larger contrastive loss weight (10) with fewer labels (ablation)CoMatch · ImageNet (1% labels, 100 epochs)top-1 accuracy (%)Improved

Later papers that compared with this method

15 later accepted papers report CoMatch as a baseline in their results tables; 11 of them report a better number on most of the results they share with it.

7 more not listed.

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy