Research map / Vision-language and multimodal / Human and embodied modeling
MDAN: Multi-Level Dependent Attention Network for Visual Emotion Analysis
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 MDAN with hierarchical global learning, level-wise local learning, and attention modules instead of the strongest listed baseline | Visual emotion classification · WEBEmo-25 | accuracy: 33.01 -> 34.28 | Helped |
| Combine global and local predictions rather than using either alone (ablation) | MDAN · Emotion-6 | accuracy: 60.73 and 58.83 -> 61.66 | Helped |
| Combine global and local predictions rather than using either alone (ablation) | MDAN · Abstract-2 | accuracy: 84.15 and 80.64 -> 83.80 | Mixed |
| Use feature fusion in the affective-semantic mapping (ablation) | MDAN local learning · WEBEmo-25 | accuracy: 31.23 -> 31.85 | Helped |
| Map fused low-semantic-level features to higher affective levels (mapping e rather than f) (ablation) | MDAN local learning · WEBEmo-25 | accuracy: 31.68 -> 31.85 | Helped |
| Disentangle the hierarchy from lower affective levels toward higher levels (ablation) | MDAN local learning · WEBEmo-25 | accuracy: 23.17 -> 31.23 | Helped |
| Add all four Q, K, V, and O pointwise convolutions to attention (ablation) | MDAN · WEBEmo-25 | accuracy: 32.78 -> 32.74 | Hurt |
| Add MHCCA with Q and O layers, without K and V pointwise convolutions (ablation) | MDAN · WEBEmo-25 | accuracy: 32.78 -> 33.75 | Helped |
| Add UpsampleAdd to MHCCA features (ablation) | MDAN · WEBEmo-25 | accuracy: 33.13 -> 33.75 | Helped |
| Use both mean and max pooling in L-CAM instead of either alone (ablation) | MDAN · WEBEmo-25 | accuracy: 34.04 and 34.17 -> 34.28 | Helped |
| Add MHCCA to the model with L-CAM (ablation) | MDAN · WEBEmo-25 | accuracy: 33.48 -> 34.28 | Helped |
| Increase MHCCA head count from 1 to 2 (ablation) | MDAN · FI-2 | accuracy: 90.23 -> 90.83 | Helped |
Closest papers by meaning
- Multiple Feature Refining Network for Visual Emotion Distribution Learning (AAAI 2025 · similarity 0.88)
- AffectGPT: A New Dataset, Model, and Benchmark for Emotion Understanding with Multimodal Large Language Models (ICML 2025 · similarity 0.87)
- Bridge Then Begin Anew: Generating Target-Relevant Intermediate Model for Source-Free Visual Emotion Adaptation (AAAI 2025 · similarity 0.87)
- Customizing Visual Emotion Evaluation for MLLMs: An Open-vocabulary, Multifaceted, and Scalable Approach (ICLR 2026 · similarity 0.86)
- Catch Your Emotion: Sharpening Emotion Perception in Multimodal Large Language Models (ICML 2025 · similarity 0.86)
- Knowledge-Aligned Counterfactual-Enhancement Diffusion Perception for Unsupervised Cross-Domain Visual Emotion Recognition (CVPR 2025 · similarity 0.85)
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.