AI Paper Atlas

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

Fine-Grained Visual Classification via Progressive Multi-Granularity Training of Jigsaw Patches

Ruoyi Du, Dongliang Chang, Ayan Kumar Bhunia, Jiyang Xie, Zhanyu Ma, Yi-Zhe Song, Jun Guo

ECCV 2020 · Label-efficient and robust learning › Recognition and long-tail · approach: Correspondence and causal representation learning · 433 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 jigsaw puzzle generator to progressive training (S=3) (ablation)PMG · CUB-200-2011 (ResNet50)combined accuracy: 88.7 -> 89.6Improved
add jigsaw puzzle generator to progressive training (S=1) (ablation)PMG · CUB-200-2011 (ResNet50)combined accuracy: 86.5 -> 86.9Improved
add jigsaw puzzle generator to progressive training (S=4) (ablation)PMG · CUB-200-2011 (ResNet50)combined accuracy: 88.5 -> 88.5No clear effect
add jigsaw puzzle generator to progressive training (S=5) (ablation)PMG · CUB-200-2011 (ResNet50)combined accuracy: 88.3 -> 87.7Hurt
increase number of supervised stages S from 1 to 3 (no jigsaw) (ablation)PMG · CUB-200-2011 (ResNet50)combined accuracy: 86.5 -> 88.7Improved
increase number of supervised stages S from 3 to 4 (no jigsaw) (ablation)PMG · CUB-200-2011 (ResNet50)combined accuracy: 88.7 -> 88.5Hurt
increase number of supervised stages S from 3 to 5 with jigsaw (ablation)PMG · CUB-200-2011 (ResNet50)combined accuracy: 89.6 -> 87.7Hurt
train progressively instead of back-propagating four outputs in one step (ablation)PMG · CUB-200-2011 (ResNet50)combined accuracy: 89.1 -> 89.6Improved
fuse granularities in one progressive network instead of four separate networks (ablation)PMG · CUB-200-2011 (ResNet50)combined accuracy: 87.7 -> 89.6Improved
raise beta from 1/3 to 2 with alpha=1 (ablation)PMG · CUB-200-2011 (ResNet50)combined accuracy: 88.6 -> 89.6Improved
raise beta from 2 to 3 with alpha=1 (ablation)PMG · CUB-200-2011 (ResNet50)combined accuracy: 89.6 -> 89.1Hurt
use combined outputs C2 instead of concat output C1 (ResNet50)PMG · CUB-200-2011accuracy: 88.9 -> 89.6Improved

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