AI Paper Atlas

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

Stochastic Partial Swap: Enhanced Model Generalization and Interpretability for Fine-Grained Recognition

Shaoli Huang, Xinchao Wang, Dacheng Tao

ICCV 2021 · Label-efficient and robust learning › Recognition and long-tail · approach: Correspondence and causal representation learning · 74 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
add Stochastic Partial Swap to mid-level branch (ablation)M-baseline (ResNet-50) · CUB-200-2011accuracy (%): 82.87 -> 87.29Improved
add Stochastic Partial Swap to mid-level branch (ablation)M-baseline (ResNet-50) · Stanford Carsaccuracy (%): 90.42 -> 94.35Improved
add Stochastic Partial Swap to mid-level branch (ablation)M-baseline (ResNet-50) · FGVC Aircraftaccuracy (%): 89.61 -> 92.31Improved
add Stochastic Partial Swap to mid-level branch (ablation)M-baseline (ResNet-50) · NABirdsaccuracy (%): 82.85 -> 86.11Improved
add Stochastic Partial Swap to mid-level branch (ablation)M-baseline (InceptionV3) · CUB-200-2011accuracy (%): 84.55 -> 86.93Improved
use SPS instead of Dropout on mid-level branchM-baseline + Dropout (p=0.5) · CUB-200-2011accuracy (%): 85.48 -> 87.29Improved
use SPS instead of Cutmix on mid-level branchM-baseline + Cutmix · CUB-200-2011accuracy (%): 85.71 -> 87.29Improved
sample swap ratio from U(0.3, 0.5) instead of fixed 0.4 (ablation)SPS · CUB-200-2011accuracy (%): 86.65 -> 87.29Improved
add SPS to promote more regions in prediction (ablation)M-baseline · CUB-200-2011 train setregions contributing 99% of prediction: 35.73 -> 56.74Improved
fuse SPS mid-level branch with high-level branchH-baseline (ResNet-50) · CUB-200-2011accuracy (%): 85.15 -> 88.42Improved
fuse SPS branch instead of plain mid-level baseline branchH-baseline + M-baseline · CUB-200-2011accuracy (%): 87.32 -> 88.42Improved
use two SPS mid-level branches instead of one (ablation)H-baseline + SPS · CUB-200-2011accuracy (%): 88.42 -> 88.70Improved

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