Research map / Vision-language and multimodal / Multimodal LLMs
Contrastive-Guided Self-Supervised Latent Visual Reasoning for Hallucination Mitigation
Aoxue Dai, Ningning Wang
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add RL stage with contrastive-guided GRPO after SFT (ablation) | CoLVR-3B · POPE | F1: 84.4 -> 85.9 | Improved |
| add RL stage with contrastive-guided GRPO after SFT (ablation) | CoLVR-3B · Hallbench | accuracy: 40.8 -> 42.6 | Improved |
| add RL stage with contrastive-guided GRPO after SFT (ablation) | CoLVR-3B · MMHal | score: 4.46 -> 4.78 | Improved |
| add RL stage with contrastive-guided GRPO after SFT (ablation) | CoLVR-3B · CHAIR | CHAIRs: 38.6 -> 30.8 | Improved |
| add SFT-only latent training to base model (ablation) | Qwen2.5-VL-3B · CHAIR | F1: 79.0 -> 78.3 | No clear effect |
| add latent-only backpropagation (stop-gradient) (ablation) | CoLVR-3B · POPE | accuracy: 84.6 -> 88.2 | Improved |
| add latent-only backpropagation (stop-gradient) (ablation) | CoLVR-3B · MMHal | score: 4.28 -> 4.78 | Improved |
| use Gaussian-noise contrast instead of blank image (ablation) | CoLVR-3B · CHAIR | CHAIRs: 36.9 -> 30.8 | Improved |
| use contrastive saliency instead of original-image attention (ablation) | CoLVR-3B · CHAIR | CHAIRs: 33.3 -> 30.8 | Improved |
| use correct contrast direction instead of reverse contrast (ablation) | CoLVR-3B · CHAIR | CHAIRs: 46.9 -> 30.8 | Improved |
| vary latent length K from 1 to 32 (ablation) | CoLVR-3B · CHAIR | CHAIRs and F1 | Mixed |
| use CoLVR self-supervised instead of Monet supervised latent reasoning | CoLVR-7B vs Monet-7B · CHAIR | CHAIRs: 34.1 -> 20.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- Deepeyes-7B — DeepEyes: Incentivizing "Thinking with Images" via Reinforcement Learning (ICLR 2026) · better on 3 of 3 shared results
- Monet-7B — Monet: Reasoning in Latent Visual Space Beyond Image and Language (CVPR 2026) · better on 3 of 3 shared results
- Pixel-Reasoner-7B — Pixel Reasoner: Incentivizing Pixel Space Reasoning via Curiosity-Driven Reinforcement Learning (NeurIPS 2025) · better on 2 of 2 shared results
- HALC — HALC: Object Hallucination Reduction via Adaptive Focal-Contrast Decoding (ICML 2024) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- See It, Say It, Sorted: An Iterative Training-Free Framework for Visually-Grounded Multimodal Reasoning in LVLMs (CVPR 2026 · similarity 0.91)
- VES-RFT: Rewarding Visual Evidence Sensitivity to Mitigate Hallucinations in Large Vision-Language Models (CVPR 2026 · similarity 0.91)
- Look Carefully: Adaptive Visual Reinforcements in Multimodal Large Language Models for Hallucination Mitigation (ICLR 2026 · similarity 0.91)
- Hallucination-aware Intermediate Representation Editing in Large Vision-Lanugage Models (ICLR 2026 · similarity 0.90)
- Dynamic Multimodal Activation Steering for Hallucination Mitigation in Large Vision-Language Models (ICLR 2026 · similarity 0.90)
- Understanding the Role of Hallucination in Reinforcement Post-Training of Multimodal Reasoning Models (CVPR 2026 · similarity 0.90)
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.
