AI Paper Atlas

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

Cross-X Learning for Fine-Grained Visual Categorization

Wei Luo, Xitong Yang, Xianjie Mo, Yuheng Lu, Larry S. Davis, Jun Li, Jian Yang, Ser-Nam Lim

ICCV 2019 · Label-efficient and robust learning › Recognition and long-tail · approach: Correspondence and causal representation learning · 298 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 GMP instead of GAP on mid-level features (with CL, no merged maps) (ablation)Cross-X (SENet-50) · NABirdstop-1 accuracy: 76.3 -> 81.7Improved
use GMP instead of GAP on mid-level features (with CL, no merged maps) (ablation)Cross-X (SENet-50) · Stanford Carstop-1 accuracy: 90.4 -> 93.0Improved
use GMP instead of GAP on mid-level features (with CL, no merged maps) (ablation)Cross-X (SENet-50) · Stanford Dogstop-1 accuracy: 87.3 -> 83.0Hurt
use GAP instead of GMP on mid-level features (with merged maps, no CL) (ablation)Cross-X (SENet-50) · Stanford Carstop-1 accuracy: 91.9 -> 93.8Improved
use GAP instead of GMP on mid-level features (with merged maps, no CL) (ablation)Cross-X (SENet-50) · CUB-Birdstop-1 accuracy: 84.2 -> 84.7No clear effect
add C3S regularizer to OSME network (ablation)OSME on SENet-50 · five FGVC benchmarkstop-1 accuracyImproved
add CL regularizer to the C3S+GMP network (ablation)OSME + C3S + GMP · five FGVC benchmarkstop-1 accuracyImproved
combine mid- and high-level features without a CL constraint (ablation)OSME + C3S · five FGVC benchmarkstop-1 accuracyHurt
add merged (FPN) feature maps (ablation)Cross-X (SENet-50) · five FGVC benchmarkstop-1 accuracyImproved
add CL when merged feature maps are used (ablation)Cross-X (SENet-50) · five FGVC benchmarkstop-1 accuracyImproved
replace metric-learning loss of MAMC-CNN with C3S and CL (Cross-X vs MAMC-CNN)Cross-X (ResNet) · Stanford Carstop-1 accuracy: 93.0 -> 94.6Improved
Cross-X ResNet-50 versus MaxEnt-CNN with DenseNet-161Cross-X (ResNet) · NABirdstop-1 accuracy: 83.0 -> 86.2Improved

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