AI Paper Atlas

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

Learning Deep Bilinear Transformation for Fine-grained Image Representation

Heliang Zheng, Jianlong Fu, Zheng-Jun Zha, Jiebo Luo

NeurIPS 2019 · Label-efficient and robust learning › Recognition and long-tail · approach: Correspondence and causal representation learning · 51 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 DBT blocks to ResNet-50ResNet-50 · Stanford-Caraccuracy: 92.8 -> 94.1Improved
Add DBT blocks to ResNet-50ResNet-50 · FGVC-Aircraftaccuracy: 90.0 -> 91.2Improved
Add DBT blocks to ResNet-50ResNet-50 · CUB-200-2011accuracy: 88.1 -> 87.5Hurt
Add DBT blocks to ResNet-50ResNet-50 · iNaturalist-2017accuracy: 59.9% -> 62.0%Improved
Add semantic grouping constraint during pre-training (ablation)DBTNet-50 · CUB-200-2011, 224 × 224accuracy: 79.8 -> 84.6Improved
Increase pre-training semantic grouping weight from 3e-4 to 3e-3 (ablation)DBTNet-50 · CUB-200-2011, 224 × 224accuracy: 84.6 -> 83.1Hurt
Use fine-tuning semantic grouping weight 1e-5 along with pre-training weight 3e-4 (ablation)DBTNet-50 · CUB-200-2011, 224 × 224accuracy: 84.6 -> 85.1Improved
Add group index encoding with t=1.5 (ablation)DBTNet-50 · CUB-200-2011, 224 × 224accuracy: 84.5 -> 85.1Improved
Use group index encoding with t=10 instead of t=1.5 (ablation)DBTNet-50 · CUB-200-2011, 224 × 224accuracy: 85.1 -> 84.5Hurt
Add shortcut connection in Stage V (ablation)DBTNet-50 · CUB-200-2011, 224 × 224, Stage Vaccuracy: 84.6 -> 84.8Improved
Add shortcut connection in Stages V+IV (ablation)DBTNet-50 · CUB-200-2011, 224 × 224, Stages V+IVaccuracy: 83.1 -> 85.1Improved
Add DBT in Stages V+IV in addition to the last layer (ablation)DBTNet-50 · CUB-200-2011, 224 × 224accuracy: 84.4 -> 85.1Improved

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