AI Paper Atlas

Research map / Vision-language and multimodal / Human and embodied modeling

The Curious Layperson: Fine-Grained Image Recognition Without Expert Labels

Subhabrata Choudhury, Iro Laina, Christian Rupprecht, Andrea Vedaldi

IJCV 2023 · Vision-language and multimodal › Human and embodied modeling · approach: Composed and text-based image retrieval · 14 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 score distribution regularizer R(B) and neutral class (ablation)FGSM w/ ensemble · CUB-200top-1 accuracy: 5.9 -> 7.9Improved
add score distribution regularizer R(B) (2-class) (ablation)FGSM w/ ensemble · CUB-200top-1 accuracy: 5.9 -> 7.4Improved
add neutral class on top of regularizer (ablation)FGSM + R [2-cls] · CUB-200top-1 accuracy: 7.4 -> 7.9Improved
add neutral class on top of regularizer (ablation)FGSM + R [2-cls] · CUB-200mean rank (lower better): 29.9 -> 31.9Hurt
replace cosine similarity with learned FGSM matching head (ablation)FGSM · CUB-200top-1 accuracy: 4.5 -> 5.9Improved
use SAT+AoANet caption ensemble instead of SAT alone (ablation)FGSM · CUB-200top-1 accuracy: 4.3 -> 5.9Improved
use ensemble instead of AoANet alone (ablation)FGSM · CUB-200top-5 accuracy: 20.8 -> 20.0Mixed
use ground-truth human descriptions instead of generated captionsFGSM · CUB-200top-1 accuracy: 5.9 -> 11.9Improved
use FGSM Siamese scoring instead of cross-encoder RoBERTadocument scoring · CUB-200inference time per image: 347 sec -> 0.55 secImproved
use captions as intermediary instead of DSCMR trained on image-document pairsours · CUB-200 (ZSL split)top-1 accuracy: 13.5 -> 20.9Improved
remove class names from documentsCLIP · CUB-200 (ZSL split)retrieval accuracyHurt

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