AI Paper Atlas

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

Causal Transportability for Visual Recognition

Chengzhi Mao, Kevin Xia, James Wang, Hao Wang, Junfeng Yang, Elias Bareinboim, Carl Vondrick

CVPR 2022 · Label-efficient and robust learning › Recognition and long-tail · approach: Correspondence and causal representation learning · 28 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
estimate the causal effect using neural representations and marginalize over sampled imagesvisual classifiers · CMNIST OODaccuracy: 29.6% -> 51.4%Improved
use a single forward pass at inference instead of marginalizing over sampled representations and images (ablation)Causal-Transportability Model · CMNIST OODaccuracy: 38.8% -> 51.4%Hurt
estimate the causal effect using neural representations and marginalize over sampled imagesWaterBird classifier with ResNet50 representation · WaterBird OODaccuracy: 52.0% -> 77.9%Improved
use a single forward pass at inference instead of marginalizing over sampled representations and images (ablation)Causal-Transportability Model · WaterBird OODaccuracy: 71.6% -> 77.9%Hurt
estimate the causal effect using neural representations and marginalize over sampled imagesImageNet-9 classifier with Moco-v2 representation · ImageNet-9 Adversarial Backgrounds OODaccuracy: 14.59% -> 18.02%Improved
use a single forward pass at inference instead of marginalizing over sampled representations and images (ablation)Causal-Transportability Model with Moco-v2 representation · ImageNet-9 Adversarial Backgrounds OODaccuracy: 17.04% -> 18.02%Hurt
estimate the causal effect using neural representations and marginalize over sampled imagesImageNet-9 classifier with SWAV representation · ImageNet-9 Adversarial Backgrounds OODaccuracy: 20.00% -> 20.42%Improved
use a single forward pass at inference instead of marginalizing over sampled representations and images (ablation)Causal-Transportability Model with SWAV representation · ImageNet-9 Adversarial Backgrounds OODaccuracy: 20.25% -> 20.42%Hurt
estimate the causal effect using neural representations and marginalize over sampled imagesImageNet-9 classifier with SimCLR representation · ImageNet-9 Adversarial Backgrounds OODaccuracy: 27.73% -> 29.41%Improved
use a single forward pass at inference instead of marginalizing over sampled representations and images (ablation)Causal-Transportability Model with SimCLR representation · ImageNet-9 Adversarial Backgrounds OODaccuracy: 28.44% -> 29.41%Hurt
estimate the causal effect using neural representations and marginalize over sampled imagesImageNet classifiers with contrastive representations · ImageNet-Rendition and ImageNet-SketchaccuracyMixed
increase the number of sampled images X' at inferenceCausal-Transportability Model · CMNIST, WaterBird, ImageNet-Rendition, and ImageNet-SketchaccuracyImproved

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