AI Paper Atlas

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

Group-Based Distinctive Image Captioning with Memory Difference Encoding and Attention

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

IJCV 2024 · Vision-language and multimodal › Human and embodied modeling · approach: Neural image captioning · 3 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 DifDisCap to TransformerTransformer · MSCOCO testDisWordRate (%): 16.8 -> 19.7Improved
add DifDisCap to TransformerTransformer · MSCOCO testCIDErBtw: 74.8 -> 70.7Improved
add DifDisCap to Transformer (accuracy cost)Transformer · MSCOCO testCIDEr: 111.7 -> 107.1Hurt
add DifDisCap to M2Transformer+SCSTM2Transformer+SCST · MSCOCO testDisWordRate (%): 17.3 -> 18.7Improved
add DifDisCap to M2Transformer+SCST (accuracy cost)M2Transformer+SCST · MSCOCO testCIDEr: 128.9 -> 127.2Hurt
add image-group-based training (ablation)M2Transformer · MSCOCO testDisWordRate (%): 16.4 -> 16.9Improved
add weighted distinctive loss (WeiDisLoss) (ablation)M2Transformer+SCST + ImageGroup · MSCOCO testDisWordRate (%): 17.4 -> 18.3Improved
add group-based differential memory attention (GDMA) (ablation)M2Transformer + ImageGroup + WeiDisLoss · MSCOCO testDisWordRate (%): 17.3 -> 18.6Improved
add Indicated Training (ablation)M2Transformer + GDMA · MSCOCO testDisWordRate (%): 18.6 -> 18.8Improved
use RegionCLIP region features instead of Bottom-up (ablation)M2Transformer+SCST+DifDisCap · MSCOCO testDisWordRate (%): 18.7 -> 18.9Improved
use ViLBERT region features instead of Bottom-up (ablation)M2Transformer+SCST+DifDisCap · MSCOCO testDisWordRate (%): 18.7 -> 18.3Hurt
raise group size K from 1 to 5 (ablation)M2Transformer+SCST+DifDisCap · MSCOCO testDisWordRate (%): 17.6 -> 18.7Mixed

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