Research map / Vision-language and multimodal / Human and embodied modeling
Learning Visual Emotion Representations From Web Data
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 |
|---|---|---|---|
| Train ResNet-50 features on StockEmotion emotion labels | EmotionNet · DeepEmotion, UnBiasedEmotion, SE30K8, EMOTIC-B, EMOTIC-I | benchmark score: 58.30 -> 65.81; 60.26 -> 81.45; 52.52 -> 69.78; 24.34 -> 29.24; 26.03 -> 30.96 | Helped |
| Train ResNet-50 features on StockEmotion emotion labels | EmotionNet · AffectNet | accuracy: 40.17 -> 53.43 | Hurt |
| Train EmotionNet with extra SE30K8 human annotations (ablation) | EmotionNet · five emotion benchmarks | benchmark score: 65.81 -> 65.53; 81.45 -> 81.45; 53.43 -> 53.69; 29.24 -> 28.98; 30.96 -> 30.99 | Mixed |
| Add the text-predicted soft-label classification loss (ablation) | EmotionNet · five emotion benchmarks | benchmark score: 64.76 -> 65.81; 80.13 -> 81.45; 52.66 -> 53.43; 28.61 -> 29.24; 30.66 -> 30.96 | Helped |
| Add the joint text-visual embedding loss (ablation) | EmotionNet · five emotion benchmarks | benchmark score: 65.85 -> 65.81; 80.29 -> 81.45; 52.71 -> 53.43; 28.74 -> 29.24; 30.83 -> 30.96 | Mixed |
| Add both soft-label and joint embedding losses (ablation) | EmotionNet · five emotion benchmarks | benchmark score: 65.29 -> 65.81; 78.98 -> 81.45; 52.51 -> 53.43; 28.58 -> 29.24; 30.52 -> 30.96 | Helped |
| Use 690 fine-grained emotion categories instead of 8 (ablation) | EmotionNet · five emotion benchmarks | benchmark score: 64.20 -> 65.29; 78.96 -> 78.98; 45.57 -> 52.51; 28.13 -> 28.58; 29.54 -> 30.52 | Helped |
| Use 690 emotion-related labels rather than 30K keywords (ablation) | EmotionNet · five emotion benchmarks | benchmark score: 63.41 -> 65.29; 74.54 -> 78.98; 46.57 -> 52.51; 27.60 -> 28.58; 28.96 -> 30.52 | Helped |
| Train on more StockEmotion images (ablation) | EmotionNet · UnBiasedEmotion | accuracy: 52.45; 66.34; 72.24; 76.63; 78.98 | Helped |
| Use EmotionNet for zero-shot recognition rather than the supervised SOTA method | EMOTIC · EMOTIC-B and EMOTIC-I | mAP: 25.44 -> 23.29; 22.48 -> 24.24 | Mixed |
Closest papers by meaning
- FindingEmo: An Image Dataset for Emotion Recognition in the Wild (NeurIPS 2024 · similarity 0.89)
- Bridge Then Begin Anew: Generating Target-Relevant Intermediate Model for Source-Free Visual Emotion Adaptation (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)
- EmoVIT: Revolutionizing Emotion Insights with Visual Instruction Tuning (CVPR 2024 · similarity 0.87)
- Multiple Feature Refining Network for Visual Emotion Distribution Learning (AAAI 2025 · similarity 0.86)
- Customizing Visual Emotion Evaluation for MLLMs: An Open-vocabulary, Multifaceted, and Scalable Approach (ICLR 2026 · similarity 0.86)
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.