AI Paper Atlas

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

On Distinctive Image Captioning via Comparing and Reweighting

Jiuniu Wang, Wenjia Xu, Qingzhong Wang, Antoni B. Chan

TPAMI 2022 · Vision-language and multimodal › Human and embodied modeling · approach: Neural image captioning · 35 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 DCR weighted XE lossTransformer · MSCOCO testCIDEr: 110.13 -> 113.98Improved
add DCR weighted XE lossTransformer · MSCOCO testCIDErBtw: 80.98 -> 75.75Improved
add DCR weighted loss and rewardTransformer+SCST · MSCOCO testCIDEr: 125.13 -> 128.53Improved
add DCR weighted loss and rewardTransformer+SCST · MSCOCO testR@1: 23.38 -> 25.84Improved
add DCR on weaker modelFC+SCST · MSCOCO testCIDEr: 104.43 -> 104.61No clear effect
add CIDErBtw reward (ablation)Transformer+SCST · MSCOCO testCIDErBtw: 86.68 -> 82.74Improved
add long-tailed weights (LTW) (ablation)Transformer+CIDErBtwReweight · MSCOCO testCIDEr: 112.44 -> 113.21Improved
add long-tailed weights (LTW) (ablation)Transformer+CIDErBtwReweight · MSCOCO testCIDErBtw: 75.35 -> 75.44No clear effect
add negative samples (NS) (ablation)Transformer+LTW · MSCOCO testCIDEr: 113.21 -> 113.98Improved
set alpha_w to 0.75 instead of 0 (ablation)Transformer+SCST+CIDErBtwStrategy · MSCOCO testCIDEr: 125.38 -> 127.78Improved
raise alpha_w to 1.25 from 1.00 (ablation)Transformer+SCST+CIDErBtwStrategy · MSCOCO testCIDEr: 127.54 -> 125.02Hurt
add LTW (F_b=5000 vs without LTW) (ablation)Transformer+DCR · MSCOCO testCIDEr: 112.82 -> 113.98Improved

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