AI Paper Atlas

Research map / Vision-language and multimodal / Cross-modal learning

CODER: Coupled Diversity-Sensitive Momentum Contrastive Learning for Image-Text Retrieval

Haoran Wang, Dongliang He, Wenhao Wu, Boyang Xia, Min Yang, Fu Li, Yunlong Yu, Zhong Ji, Errui Ding, Jingdong Wang

ECCV 2022 · Vision-language and multimodal › Cross-modal learning · approach: Probabilistic embeddings for cross-modal retrieval · 39 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
replace BTR loss with the diversity-sensitive contrastive (DCL) loss (ablation)CODER baseline · Flickr30KR@1: 78.7 -> 81.9 text, 58.6 -> 61.5 imageImproved
add coupled memory banks (M-DCL) to the DCL loss (ablation)CODER · Flickr30KR@1: 81.9 -> 82.6 text, 61.5 -> 62.6 imageImproved
add concept-level DCL loss (ablation)CODER · Flickr30KR@1: 82.6 -> 82.8 text, 62.6 -> 62.8 imageImproved
add pseudo-clustering label prediction (PGC) loss (ablation)CODER · Flickr30KR@1: 82.8 -> 83.2 text, 62.8 -> 63.1 imageImproved
add explicit diversity estimation to implicit DCL (ablation)DCL loss · Flickr30KR@1: 80.3 -> 81.5 text, 60.2 -> 61.2 imageImproved
use memory bank items in explicit diversity estimation (ablation)DCL loss · Flickr30KR@1: 81.5 -> 81.9 text, 61.2 -> 61.5 imageImproved
combine M-DCL with instance-level DCL loss (ablation)CODER · Flickr30KR@1: 81.9 -> 83.2 text, 61.5 -> 63.1 imageImproved
add M-DCL when batch size shrinks from 128 to 32 (ablation)CODER · Flickr30Kimage/text R@1: without M-DCL 61.5 -> 59.3 image, 81.9 -> 79.3 text; with M-DCL only 0.9% dropImproved
vary cluster number K in PGC loss from 5000 to 20000 (ablation)CODER · Flickr30KR@1: text R@1 82.9, 83.2, 83.2, 82.8 for K 5000 to 20000No clear effect
CODER versus best competitor SGRAFCODER · Flickr30KR@sum: 499.6 -> 520.9Improved
CODER versus best competitor SGRAFCODER · MSCOCO 1KR@1 text retrieval: 79.6 -> 82.1Improved

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