Research map / Vision-language and multimodal / Multimodal LLMs

Chain-of-Thought Guided Multi-Modal Object Re-Identification

CVPR 2026 · Vision-language and multimodal › Multimodal LLMs · approach: Chain-of-thought multimodal reasoning

Read the paperCheck your idea against 108,967 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Proposes CoT-ReID, which uses MLLM-generated chain-of-thought text to guide a DINOv3-based multi-modal (RGB/NIR/TIR) object re-identification model at early, cross-modal and decision levels.

Problem

Multi-modal object ReID (RGB, NIR, TIR and text) mostly uses static descriptive captions and ignores the logical hierarchy among semantic attributes; the paper asks whether CoT reasoning text can supply that structure.

Approach

Generates CoT reasoning chains and attribute captions per modality with Qwen-VL, then uses the CoT text in three places: as reverse embeddings in DINOv3 register tokens (CRE), as a semantic anchor and protective margin for cross-modal contrastive consistency (CT-CMC, building on InfoBridge), and as decision features (CTF).

Key results

• On WMVeID863, CoT-ReID reaches 74.7 mAP and 82.0 R-1, versus 70.1 mAP for the DINOv3 baseline.
• On MSVR310, CoT-ReID reaches 71.7 mAP, improving over DeMo (DINOv3) by 3.0 mAP and IDEA (DINOv3) by 4.7 mAP.
• On RGBNT201, CoT-ReID reaches 83.3 mAP and 86.1 R-1, versus 81.3 mAP for IDEA (DINOv3).

Limitations

• Relies on API-generated Qwen-VL text whose quality and cost are not analysed in the main text.
• Reported gains over the DINOv3 baseline are modest on some datasets (e.g., RGBNT100 R-1 98.5 vs 99.3 range) and the paper does not separate which reasoning signals matter.
• Hyperparameters are tuned per dataset with differing learning rates, which may limit generality.

Experimental setup

Datasets: MSVR310 RGBNT100 WMVeID863 RGBNT201

Benchmarks: RGBNT100 WMVeID863 MSVR310 RGBNT201

Metrics: mAP CMC Rank-1 Rank-5 Rank-10

Baselines: HAMNet CCNet EDITOR TOP-ReID DeMo IDEA ICPL Mambapro 3M loss DINOv3

Builds on: DINOv3 CLIP InfoBridge IDEA DeMo

Closest papers by meaning

• Uni-CoT: Towards Unified Chain-of-Thought Reasoning Across Text and Vision (ICLR 2026 · similarity 0.91)
• TVI-CoT: Text-Visual Interleaved Chain-of-Thought Reasoning for Multimodal Understanding (ICML 2026 · similarity 0.90)
• Multi-Modal Latent Space Learning for Chain-of-Thought Reasoning in Language Models (AAAI 2024 · similarity 0.90)
• Visual CoT: Advancing Multi-Modal Language Models with a Comprehensive Dataset and Benchmark for Chain-of-Thought Reasoning (NeurIPS 2024 · similarity 0.89)
• CoMT: A Novel Benchmark for Chain of Multi-modal Thought on Large Vision-Language Models (AAAI 2025 · similarity 0.89)
• Rationale-Enhanced Decoding for Multi-modal Chain-of-Thought (CVPR 2026 · similarity 0.89)