AI Paper Atlas

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

Cross-modal Identity Mapping: Minimizing Information Loss in Modality Conversion via Reinforcement Learning

Haonan Jia, Shichao Dong, Xin Dong, Zenghui Sun, Jin Wang, Jinsong Lan, Xiaoyong Zhu, Bo Zheng, Kaifu Zhang

CVPR 2026 · Vision-language and multimodal › Cross-modal learning · approach: Contrastive vision-language alignment · 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
train captioner with CIM reward (GRC + QIR) via GRPOQwen2.5-VL-7B · COCO-LN500Relations QA: 23.76 -> 44.15Improved
train captioner with CIM rewardQwen2.5-VL-7B · COCO-LN500Attributes F1: 46.25 -> 58.51Improved
train captioner with CIM rewardQwen2.5-VL-7B · COCO-LN500Objects precision: 82.35 -> 80.31Hurt
train captioner with CIM rewardLLaVA1.5-7B · COCO-LN500CAPTURE: 44.75 -> 48.62Improved
add QIR-only reward (vs base) (ablation)Qwen2.5-VL-7B · COCO-LN500CAPTURE/Objects F1/Attributes F1/Relations QA: +4.0, +11.0, +10.8, +17.4Improved
add GRC-only reward (vs base) (ablation)Qwen2.5-VL-7B · COCO-LN500CAPTURE/Objects F1/Attributes F1/Relations QA: +2.9, +3.6, +5.1, +0.9Improved
combine QIR and GRC rewards (ablation)CIM · COCO-LN500CAPTURE and sub-metricsImproved
CIM on top of SFT model vs SC-CaptionerQwen2-VL-7B SFT · COCO-LN500CAPTURE: 47.51 -> 48.61Improved
CIM on top of SFT model vs SC-CaptionerLLaVA1.5-7B SFT · COCO-LN500 and DOCCI500CAPTURE: +0.4 avgNo clear effect
scale retrieval corpus (RefinedCaps -> +DenseFusion-1M) (ablation)CIM · COCO-LN500Objects F1 / Attributes F1 / Relations QA: +1.1/+0.6/+1.3 then +0.4/+0.3/+1.7Improved
swap retrieval image/text encoders (ablation)CIM · COCO-LN500CAPTURE: 48.17 - 48.99No clear effect
CIM reward vs CLIP-RL and Self Retrieval baselinesQwen2.5-VL-7B · COCO-LN500CAPTURE: 46.71 -> 48.93Improved

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