Research map / Video / Video-language understanding
StimuVAR: Spatiotemporal Stimuli-Aware Video Affective Reasoning with Multimodal Large Language Models
Yuxiang Guo, Faizan Siddiqui, Yang Zhao, Rama Chellappa, Shao-Yuan Lo
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 event-driven frame sampling (ablation) | vanilla model · VCE | Top-1/Top-3 accuracy: 44.5 -> 46.0 Top-1; 68.4 -> 70.9 Top-3 | Improved |
| add VAR visual instruction tuning (ablation) | model with event-driven sampling · VCE | Top-1/Top-3 accuracy: 46.0 -> 47.3 Top-1; 70.9 -> 72.5 Top-3 | Improved |
| add emotion-triggered tube selection (ablation) | model with sampling and instruction tuning · VCE | Top-1/Top-3 accuracy: 47.3 -> 48.3 Top-1; 72.5 -> 73.5 Top-3 | Improved |
| tube selection instead of Valley V3 aggregation (ablation) | StimuVAR · VCE | Top-3 accuracy: 72.9 -> 73.5 | Improved |
| tube selection instead of TS Selection (ablation) | StimuVAR · VCE | Top-3 accuracy: 72.5 -> 73.5 | Improved |
| event-driven instead of uniform frame sampling (ablation) | StimuVAR · VCE surprise subset | Top-3 accuracy: 67.6 -> 72.9 | Improved |
| event-driven instead of uniform frame sampling (ablation) | StimuVAR · VCE entire test set | Top-3 accuracy: 69.7 -> 73.5 | Improved |
| event-driven instead of AKS frame selection (ablation) | StimuVAR · VCE entire test set | Top-3 accuracy: 71.6 -> 73.5 | Improved |
| event-driven instead of uniform sampling (ablation) | Chat-UniVi (fine-tuned) · VCE surprise subset | Top-3 accuracy: 57.9 -> 63.2 | Improved |
| increase sampled frames from N=3 to N=6 (ablation) | StimuVAR · VCE | Top-3 accuracy: 69.9 -> 73.5 | Improved |
| select top K=4 tubes instead of K=1 (ablation) | StimuVAR · VCE | Top-3 accuracy: 71.8 -> 73.5 | Improved |
| use Llama3-8b instead of Llama2-7b (ablation) | StimuVAR · VCE | Top-3 accuracy: 73.5 -> 74.0 | Improved |
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.
- VideoMAE — VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training (NeurIPS 2022) · 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
- VidEmo: Affective-Tree Reasoning for Emotion-Centric Video Foundation Models (NeurIPS 2025 · similarity 0.90)
- Customizing Visual Emotion Evaluation for MLLMs: An Open-vocabulary, Multifaceted, and Scalable Approach (ICLR 2026 · similarity 0.89)
- AffectGPT: A New Dataset, Model, and Benchmark for Emotion Understanding with Multimodal Large Language Models (ICML 2025 · similarity 0.88)
- MME-Emotion: A Holistic Evaluation Benchmark for Emotional Intelligence in Multimodal Large Language Models (ICLR 2026 · similarity 0.87)
- Look, Remember and Reason: Grounded Reasoning in Videos with Language Models (ICLR 2024 · similarity 0.87)
- Catch Your Emotion: Sharpening Emotion Perception in Multimodal Large Language Models (ICML 2025 · similarity 0.87)
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.
