AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

Benchmarking Representation Learning for Natural World Image Collections

Grant Van Horn, Elijah Cole, Sara Beery, Kimberly Wilber, Serge Belongie, Oisin Mac Aodha

CVPR 2021 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Invariance-based self-supervised learning · 14 citations (OpenAlex)

Check 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
pretrain with supervised labels on iNat2021 instead of ImageNetResNet50 feature extractor · NeWT (164-task mean)top-1 accuracy: 0.744 -> 0.806Improved
use iNat2021 full training data instead of iNat2021 minisupervised ResNet50 feature extractor · NeWT (164-task mean)top-1 accuracy: 0.793 -> 0.806Improved
use iNat2021 full training data instead of iNat2021 miniSimCLR feature extractor · NeWT (164-task mean)top-1 accuracy: 0.678 -> 0.689Improved
pretrain with self-supervision on iNat2021 rather than ImageNetself-supervised feature extractors · CUBtop-1 accuracy: 0.631 -> 0.365Hurt
use ImageNet SwAV rather than supervised ImageNet featuresResNet50 feature extractor · StanfordCarstop-1 accuracy: 0.479 -> 0.595Improved
use self-supervised rather than supervised ImageNet featuresfeature extractor · NeWT Gestalttop-1 accuracy: 0.859 -> 0.908Improved
use self-supervised rather than supervised ImageNet featuresfeature extractor · NeWT Appearance - Speciestop-1 accuracy: 0.725 -> 0.647Hurt
use iNat2021 mini SimCLR x4 instead of iNat2021 mini SimCLRResNet50 feature extractor · Flowers102top-1 accuracy: 0.814 -> 0.911Improved
initialize supervised training randomly instead of from ImageNet (ablation)ResNet50 trained on iNat2021 full · iNat2021 testtop-1 accuracy: 0.760 -> 0.746Hurt
initialize supervised training randomly instead of from ImageNet (ablation)ResNet50 trained on iNat2021 mini · iNat2021 testtop-1 accuracy: 0.654 -> 0.616Hurt
add a geographical prior trained on iNat2021 to the iNat2021 vision modelsupervised ResNet50 · iNat2021 testtop-1 accuracy: 0.760 -> 0.800Improved
use a geographical prior trained on full iNat2021 rather than minivision iNat2021 mini + geographical prior · iNat2021 testtop-1 accuracy: 0.722 -> 0.739Improved

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