Emotion Recognition on RAF-DB, FERPlus, AffectNet, SFEW2.0 Composite
84.8F1 Score (RAF-DB)ExpLLM*
Evaluation results
| Method | Links | ||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ExpLLM*Model Category=MLLM-based FER methods, Backbone Size=7B, Evaluation Source=reproduced from official checkpoints2026.04 | 84.8 | 54.37 | 46.86 | 43.49 | 69.33 | 57.26 | 68.52 | 84.78 | 71.23 | 69.58 | 54.79 | 46.83 | 62.36 | 0 | |
| ActFER <4B>Backbone=Qwen3VL-4B, Training=Reinforcement Learning (UC-GRPO)2026.04 | 82.72 | 59.92 | 57.66 | 51.13 | 73.89 | 67.45 | 79.28 | 89.34 | 70.39 | 70.9 | 60 | 54.71 | 63.97 | 51 | |
| UniFER-7B RLModel Category=MLLM-based FER methods, Backbone Size=7B2026.04 | 81.3 | 58.55 | 44.53 | 39.7 | 68.84 | 55.32 | 72.8 | 84.32 | 72.46 | 51.31 | 51.57 | 41.68 | 65.08 | 3.31 | |
| ActFER-SFT <4B>Backbone=Qwen3VL-4B, Training=Supervised Fine-Tuning2026.04 | 74.87 | 51.94 | 45.53 | 46.61 | 65.7 | 53.37 | 70.33 | 84.6 | 65.44 | 64.78 | 41.03 | 41.54 | 58.14 | 1.11 | |
| InternVL3.5-4B*Model Category=General-purpose MLLMs, Backbone Size=4B, Evaluation Source=reproduced from official checkpoints2026.04 | 60.69 | 37.87 | 44.51 | 33.96 | 50.4 | 43.67 | 55.56 | 81.89 | 53.47 | 55.06 | 17.44 | 35.59 | 46.68 | 3.7 | |
| LLaVA-Next-34BModel Category=General-purpose MLLMs, Backbone Size=34B2026.04 | 60.56 | 48.43 | 34.92 | 37.76 | 61.2 | 44.36 | 60.62 | 86.11 | 64 | 53.79 | 3.01 | 35.79 | 51.22 | 0.36 | |
| Qwen3VL-4B*Model Category=General-purpose MLLMs, Backbone Size=4B, Evaluation Source=reproduced from official checkpoints2026.04 | 56.13 | 47.37 | 36.12 | 38.57 | 55.27 | 44.03 | 60.12 | 81.93 | 54.01 | 56.06 | 22.97 | 23.47 | 50 | 3.65 | |
| Gemini-2.5-FlashModel Category=General-purpose MLLMs2026.04 | 55.6 | 44.95 | 45.38 | 37.2 | 61.55 | 45.47 | 62.25 | 81.53 | 59.69 | 63.43 | 42.52 | 41.82 | 48.86 | 9.14 | |
| EmoLA*Model Category=MLLM-based FER methods, Backbone Size=7B, Evaluation Source=reproduced from official checkpoints2026.04 | 53.46 | 45.29 | 35.12 | 37.91 | 62.75 | 43.19 | 55.13 | 77.47 | 47.05 | 48.67 | 28.7 | 33.5 | 38.37 | 16.62 | |
| InternVL3-8BModel Category=General-purpose MLLMs, Backbone Size=8B2026.04 | 52.69 | 43.34 | 39.33 | 37.27 | 60.59 | 44.54 | 65.08 | 85.56 | 61.57 | 40.02 | 41.35 | 38.31 | 54.41 | 14.55 | |
| Gemini-2.5-ProModel Category=General-purpose MLLMs2026.04 | 50.95 | 39.78 | 43.11 | 36.33 | 57.17 | 44.29 | 47.43 | 81.22 | 57.78 | 63.16 | 42.42 | 39.66 | 46.44 | 20.49 | |
| Qwen2.5VL-7BModel Category=General-purpose MLLMs, Backbone Size=7B2026.04 | 50.26 | 46.25 | 26.97 | 36.22 | 53.78 | 35.34 | 54.6 | 67.29 | 53.91 | 52.7 | 3.28 | 22.26 | 46.97 | 0.37 |