Research map / Vision-language and multimodal / Cross-modal learning
Rethinking Attention Reallocation for Multimodal Emotion Recognition
Yazhe Lyu, Yixiong Zou, Jinghan Hu, Yuhua Li, Ruixuan Li
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 shallow layer guidance (ablation) | AffectGPT baseline · MER-UniBench (video+audio+text) | MER-UniBench avg: 74.34 -> 76.48 | Improved |
| add residual refinement on top of shallow guidance (ablation) | AffectGPT baseline · MER-UniBench (video+audio+text) | MER-UniBench avg: 76.48 -> 78.08 | Improved |
| replace residual injection with shallow layer replacement (ablation) | Ours · MER-UniBench (video+audio+text) | MER-UniBench avg: 78.08 -> 77.68 | Hurt |
| replace shallow prior with uniform prior injection (ablation) | Ours · MER-UniBench (video+audio+text) | MER-UniBench avg: 78.08 -> 77.27 | Hurt |
| aggregate multiple shallow layers instead of single selected layer (ablation) | Ours · MER-UniBench (video+audio+text) | MER-UniBench avg: 78.08 -> 76.01 | Hurt |
| reallocate attention from generated to modality tokens | AffectGPT · MER2023 two-modality | score: 78.17 -> 79.04 | Improved |
| reallocate attention from generated to modality tokens | AffectGPT · MER2023 three-modality | score: 76.52 -> 76.52 | Hurt |
| reallocate attention from modality to generated tokens | AffectGPT · MER2023 three-modality | score: 76.52 -> 80.53 | Improved |
| add length-constrained decoding to baseline | AffectGPT · MER-UniBench (video+audio+text) | MER-UniBench avg: 74.34 -> 71.67 | Hurt |
| add DeCo-style layer-guided logit refinement | AffectGPT · MER-UniBench (video+audio+text) | MER-UniBench avg: 74.34 -> 74.33 | No clear effect |
| add our method without pre-fusion module (ablation) | AffectGPT w/o fused token · MER-UniBench w/o fused token | MER-UniBench avg: 73.47 -> 74.81 | Improved |
| add DoLA | AffectGPT · MER-UniBench (video+audio+text) | MER-UniBench avg: 74.34 -> 68.38 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- AffectGPT — AffectGPT: A New Dataset, Model, and Benchmark for Emotion Understanding with Multimodal Large Language Models (ICML 2025) · better on 11 of 12 shared results
- Emotion-LLaMA — Emotion-LLaMA: Multimodal Emotion Recognition and Reasoning with Instruction Tuning (NeurIPS 2024) · 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
- MODA: MOdular Duplex Attention for Multimodal Perception, Cognition, and Emotion Understanding (ICML 2025 · similarity 0.91)
- Catch Your Emotion: Sharpening Emotion Perception in Multimodal Large Language Models (ICML 2025 · similarity 0.89)
- Multi-to-Single: Reducing Multimodal Dependency in Emotion Recognition Through Contrastive Learning (AAAI 2025 · similarity 0.88)
- AffectGPT: A New Dataset, Model, and Benchmark for Emotion Understanding with Multimodal Large Language Models (ICML 2025 · similarity 0.88)
- Emotion-Coherent Reasoning for Multimodal LLMs via Emotional Rationale Verifier (AAAI 2026 · similarity 0.87)
- Emotion-LLaMA: Multimodal Emotion Recognition and Reasoning with Instruction Tuning (NeurIPS 2024 · 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.
