Research map / Video / Video-language understanding
SEASON: Mitigating Temporal Hallucination in Video Large Language Models via Self-Diagnostic Contrastive Decoding
Chang-Hsun Wu, Kai-Po Chang, Yu-Yang Sheng, Hung-Kai Chung, Kuei-Chun Wang, Yu-Chiang Frank 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 |
|---|---|---|---|
| replace temporal homogenization with shuffled-frame negative (ablation) | SEASON · temporal benchmark AVG | accuracy: 61.4 -> 55.0 | Hurt |
| replace temporal homogenization with average negative (ablation) | SEASON · temporal benchmark AVG | accuracy: 61.4 -> 59.9 | Hurt |
| replace temporal homogenization with reverse negative (ablation) | SEASON · temporal benchmark AVG | accuracy: 62.0 -> 58.8 | Hurt |
| add temporal negative on top of spatial negative (full SEASON) (ablation) | SEASON · hallucination benchmarks AVG | accuracy: 63.7 -> 64.4 | Improved |
| add spatial negative to temporal negative (full SEASON) (ablation) | SEASON · hallucination benchmarks AVG | accuracy: 66.2 -> 66.5 | Improved |
| add self-diagnostic combination vs temporal negative only (ablation) | SEASON · EventHallusion | accuracy: 68.0 -> 66.0 | Hurt |
| add SEASON decoding | Qwen2.5-VL-7B · VidHalluc | accuracy: 73.4 -> 78.7 | Improved |
| add SEASON decoding | LLaVA-OV-7B · VidHalluc TSH | accuracy: 53.00 -> 77.50 | Improved |
| add SEASON decoding | LLaVA-OV-7B · VidHalluc STH | accuracy: 63.65 -> 60.29 | Hurt |
| add SEASON decoding to 72B model | LLaVA-OV-72B · VidHalluc | accuracy: 69.41 -> 71.46 | Improved |
| add SEASON decoding | LLaVA-Video-7B · TempCompass | accuracy: 69.7 -> 71.1 | Improved |
| add SEASON decoding | LLaVA-Video-7B · VideoMME | accuracy: 53.0 -> 53.4 | 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 Hallucination in VideoLLMs via Temporal-Aware Activation Engineering (NeurIPS 2025 · similarity 0.93)
- Hallucination-aware Intermediate Representation Editing in Large Vision-Lanugage Models (ICLR 2026 · similarity 0.92)
- VidHalluc: Evaluating Temporal Hallucinations in Multimodal Large Language Models for Video Understanding (CVPR 2025 · similarity 0.92)
- Generate, but Verify: Reducing Hallucination in Vision-Language Models with Retrospective Resampling (NeurIPS 2025 · similarity 0.92)
- CODE: Contrasting Self-generated Description to Combat Hallucination in Large Multi-modal Models (NeurIPS 2024 · 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.
