AI Paper Atlas

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

Leveraging Textual Compositional Reasoning for Robust Change Captioning

Kyu Ri Park, Jiyoung Park, Seong Tae Kim, Hong Joo Lee, Jung Uk Kim

AAAI 2026 · Vision-language and multimodal › Human and embodied modeling · approach: Concept-guided image captioning · 1 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 CORTEXSCORER · CLEVR-Change (total)CIDEr: 126.8 -> 128.8Improved
add CORTEXDIRL · CLEVR-Change (total)CIDEr: 125.3 -> 130.7Improved
add CORTEXDIRL · Spot-the-DiffCIDEr: 40.9 -> 49.5Improved
add CORTEXSMART · Spot-the-DiffMETEOR: 13.5 -> 12.2Hurt
add CORTEXSCORER · CLEVR-DCCIDEr: 83.7 -> 88.7Improved
add RTE text module (ablation)DIRL · CLEVR-Change (total)CIDEr: 125.3 -> 128.5Improved
add ITDA on top of RTE (ablation)DIRL · CLEVR-Change (total)CIDEr: 128.5 -> 130.7Improved
add static alignment loss Lsa (ablation)DIRL+CORTEX · CLEVR-Change (total)CIDEr: 127.9 -> 128.4Improved
add dynamic alignment loss Lda (ablation)DIRL+CORTEX · CLEVR-Change (total)CIDEr: 127.9 -> 128.9Improved
use compositional prompt instead of generic descriptions (ablation)CORTEX (DIRL) · CLEVR-Change (total)CIDEr: 129.5 -> 130.7Improved
use InternVL2 instead of LLaVA (ablation)CORTEX (DIRL) · CLEVR-Change (total)CIDEr: 130.0 -> 130.7Improved
use VLM on single images as auxiliary context instead of direct paired prediction (ablation)CORTEX (DIRL) · Spot-the-DiffCIDEr: 12.3 -> 49.5Improved

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