Research map / Vision-language and multimodal / Multimodal LLMs
VERDICT: Training-Free Step-Wise Verification of Multimodal Reasoning via Disagreement-Aware Consensus
Rohit Sinha, Kunal Tilaganji, Tanuja Ganu, Nagarajan Natarajan, Amit Sharma, Vineeth N Balasubramanian
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 |
|---|---|---|---|
| use disagreement-aware consensus verification (VERDICT) over unverified base model | Qwen2.5-VL-7B-Instruct · six benchmarks average | accuracy (%): 66.31 -> 70.15 | Improved |
| use VERDICT consensus instead of Mean aggregation of the same three agents | VERDICT · six benchmarks | accuracy (%): +0.68 to +2.57 | Improved |
| use VERDICT consensus instead of Variance filtering | VERDICT · six benchmarks average | accuracy (%): 67.65 -> 70.15 | Improved |
| use frozen consensus verifier instead of trained domain-specific critic (Sherlock) | Qwen2.5-VL-7B-Instruct · CV-Bench-3D | accuracy (%): drops 18.26 points on CV-Bench-3D | Hurt |
| add trained critic VisionSR1 to the base model | Qwen2.5-VL-7B-Instruct · CV-Bench-3D | accuracy (%): drops 22.21 points on CV-Bench-3D and 18.22 points on BLINK | Hurt |
| add intelligent ranking (selection) to the filter (ablation) | VERDICT · six benchmarks | accuracy (%): -1.17 to -2.17 when removed | Improved |
| add rejection filtering to ranking (ablation) | VERDICT · six benchmarks | accuracy (%): -0.26 to -1.08 when removed | Improved |
| use coupled consensus scores instead of raw average scores in dual criterion (ablation) | VERDICT · six benchmarks | accuracy (%): 2.59-2.95 points | Improved |
| set Visual agent stubbornness highest rather than swapping V with Contextual (ablation) | VERDICT · 3DSRBench | accuracy (%): 59.02 -> 57.26 | Improved |
| keep Logical above Contextual stubbornness (undo L-C swap) (ablation) | VERDICT · 3DSRBench | accuracy (%): 59.02 -> 58.61 | Improved |
| relax confidence threshold tau to <=0.01 (disable filter) (ablation) | VERDICT · 3DSRBench | accuracy (%): 59.02 -> 68.71 | Improved |
| relax confidence threshold tau to <=0.01 (disable filter) (ablation) | VERDICT · other five benchmarks | accuracy (%): +0.90 (AI2D) to +1.65 (MMStar) gain lost | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- LLaVA Critic — LLaVA-Critic: Learning to Evaluate Multimodal Models (CVPR 2025) · better on 1 of 1 shared result
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
- CrossCheck-Bench: Diagnosing Compositional Failures in Multimodal Conflict Resolution (AAAI 2026 · similarity 0.89)
- RFEval: Benchmarking Reasoning Faithfulness under Counterfactual Reasoning Intervention in Large Reasoning Models (ICLR 2026 · similarity 0.87)
- R-C2: Cycle-Consistent Reinforcement Learning Improves Multimodal Reasoning (CVPR 2026 · similarity 0.87)
- CARE What Fails: Contrastive Anchored-REflection for Verifiable Multimodal Reasoning (CVPR 2026 · similarity 0.87)
- Let's Think in Two Steps: Mitigating Agreement Bias in MLLMs with Self-Grounded Verification (ICLR 2026 · similarity 0.86)
- Sample Smart, Not Hard: Correctness-First Decoding for Better Reasoning in LLMs (ICLR 2026 · similarity 0.86)
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.
