AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Demystifying Unsupervised Semantic Correspondence Estimation

Mehmet Aygün, Oisin Mac Aodha

ECCV 2022 · Label-efficient and robust learning › Recognition and long-tail · approach: Correspondence and causal representation learning · 12 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 asymmetric temperatures (tau1<tau2) with distance loss instead of LEAD (ablation)LEAD, sup. CNN · SPair-71kPCK@0.1 (%): 31.7 -> 34.0Improved
use asymmetric temperatures instead of LEAD (ablation)LEAD, sup. CNN · CUBPCK@0.1 (%): 51.5 -> 60.8Improved
use asymmetric temperatures instead of LEAD (ablation)LEAD, sup. CNN · AFLWPCK@0.1 (%): 58.0 -> 63.6Improved
use asymmetric temperatures instead of LEAD (ablation)LEAD, unsup. Transformer · SPair-71kPCK@0.1 (%): 33.6 -> 32.9Hurt
change LEAD temperature (ablation)LEAD · SPair-71kPCK@0.1 (%): 31.7, 31.9, 31.4No clear effect
increase tau1,tau2 from 0.2,0.4 to 0.4,0.8 (ablation)ASYM · SPair-71kPCK@0.1 (%): 34.0 -> 29.5Hurt
increase CL temperature from 0.05 to 0.2 (ablation)CL · SPair-71kPCK@0.1 (%): 8.2 -> 26.6Improved
use MSE instead of cross-entropy for LEAD (tau=0.05) (ablation)LEAD · SPair-71kPCK@0.1 (%): 31.7 -> 31.5No clear effect
use MSE instead of cross-entropy for ASYM (ablation)ASYM · SPair-71kPCK@0.1 (%): 32.0 -> 34.0Improved
use conv3 features instead of conv4 (ablation)ResNet-50 features, no projection · SPair-71kPCK@0.1 (%): 15.8 -> 31.8Improved
use unsupervised pre-trained Transformer instead of supervised (ablation)no projection · SDogsPCK@0.1 (%): 38.0 -> 42.7Improved
penalize swap errors (PCK-dagger instead of PCK)ASYM · SPair-71kPCK-dagger vs PCK: 34.0 -> 27.2Hurt

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