Research map / Video / Video-language understanding
MACD: Model-Aware Contrastive Decoding via Counterfactual Data for Video-LLMs
Qixin Xiao, Kun Zhou
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 model-aware optimized counterfactual masks in contrastive decoding instead of random perturbation (VCD) | Video-LLM contrastive decoding · EventHallusion | Accuracy: 0.72 -> 0.82 (Qwen3-VL-2B, VCD vs MACD) | Improved |
| add gradient-based mask strength training (ablation) | MACD · EventHallusion (Qwen2.5-VL-3B) | Accuracy: 0.61 -> 0.71 | Improved |
| add frame-level mask (ablation) | MACD · EventHallusion (Qwen2.5-VL-3B) | Accuracy: 0.61 -> 0.71 | Improved |
| frame extraction (temporal subsampling) on counterfactual view without frame mask (ablation) | MACD without frame mask · EventHallusion (Qwen2.5-VL-3B) | Accuracy: 0.61 -> 0.64 | Improved |
| use masks rather than learned noise for the counterfactual (ablation) | MACD · EventHallusion (Qwen2.5-VL-3B) | Accuracy: 0.65 (global noise) / 0.67 (object noise) / 0.66 (frame noise) -> 0.71 | Improved |
| discretize mask strengths into three levels instead of continuous (ablation) | MACD · appendix Table 9 | Accuracy | Improved |
| swap YOLOv11 detector for Grounding DINO (ablation) | MACD · appendix Table 16 | Accuracy | No clear effect |
| apply MACD instead of baseline decoding on object-hallucination (video-POPE) questions | Qwen2.5-VL-3B · Video-POPE (EventHallusion subset) | Accuracy: 0.72 -> 0.85 | Improved |
| apply MACD to reduce false-yes answers on absent-object questions | Qwen2.5-VL-3B · Video-POPE (EventHallusion subset) | false-yes rate: 40.0% -> 17.0% | Improved |
| use MACD masks vs random masks of equal area (ablation) | mask construction · Human evaluation, 50 EventHallusion videos | mean score: 2.40 -> 3.10 | Improved |
| apply MACD to Qwen2-VL-7B on Video-MME | Qwen2-VL-7B · Video-MME | Accuracy: 0.60 (VCD) vs 0.59 (MACD); baseline 0.51 | Hurt |
| vary alpha and beta decoding coefficients (ablation) | MACD · EventHallusion (Qwen2.5-VL-7B) | Accuracy: 0.55 -> 0.61 -> 0.58 across alpha 2.1 to 3.8 | No clear effect |
Closest papers by meaning
- Learning to Decode Against Compositional Hallucination in Video Multimodal Large Language Models (ICML 2026 · similarity 0.93)
- Mitigating Object Hallucinations in Large Vision-Language Models through Visual Contrastive Decoding (CVPR 2024 · similarity 0.93)
- SECOND: Mitigating Perceptual Hallucination in Vision-Language Models via Selective and Contrastive Decoding (ICML 2025 · similarity 0.93)
- CODE: Contrasting Self-generated Description to Combat Hallucination in Large Multi-modal Models (NeurIPS 2024 · similarity 0.93)
- ConVis: Contrastive Decoding with Hallucination Visualization for Mitigating Hallucinations in Multimodal Large Language Models (AAAI 2025 · similarity 0.93)
- Envision, Attend, Then Respond: Counterfactual Hallucination Mitigation in Large Vision-Language Models (CVPR 2026 · similarity 0.92)
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.
