# The Failure Is in the Readout: Fine-Grained Emotion Recognition Benchmarks Measure Elicitation, Not Perception

arXiv: https://arxiv.org/abs/2610.08162 · HTML: https://mlsanity.com/paper/2610.08162/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 13 of its 61 references that are in the map.
- First posted on arXiv 6 Oct 2026.

## Builds on (most cited first, 13 of 13)

- [Qwen2.5-VL Technical Report](https://mlsanity.com/paper/2502.13923/) (2025)
- [Qwen3-VL Technical Report](https://mlsanity.com/paper/2511.21631/) (2025)
- [Gemma 3 Technical Report](https://mlsanity.com/paper/2503.19786/) (2025)
- [InternVL3.5: Advancing Open-Source Multimodal Models in Versatility, Reasoning, and Efficiency](https://mlsanity.com/paper/2508.18265/) (2025)
- [SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Features](https://mlsanity.com/paper/2502.14786/) (2025)
- [GLM-4.5V and GLM-4.1V-Thinking: Towards Versatile Multimodal Reasoning with Scalable Reinforcement Learning](https://mlsanity.com/paper/2507.01006/) (2025)
- [Are we done with ImageNet?](https://mlsanity.com/paper/2006.07159/) (2020)
- [Gemma 4 Technical Report](https://mlsanity.com/paper/2607.02770/) (2026)
- [MiniCPM-V 4.5: Cooking Efficient MLLMs via Architecture, Data, and Training Recipe](https://mlsanity.com/paper/2509.18154/) (2025)
- [MiMo-VL Technical Report](https://mlsanity.com/paper/2506.03569/) (2025)
- [Ministral 3](https://mlsanity.com/paper/2601.08584/) (2026)
- [Beyond Generative Decoding: Discriminative Hidden-State Readout from a Native Omni-Modal LLM for Multimodal Sentiment Analysis](https://mlsanity.com/paper/2606.05713/) (2026)
- [Separating Decision-Rule Misalignment from Readout-Coverage Limitations in Speech Language Models](https://mlsanity.com/paper/2608.06409/) (2026)

## Abstract (arXiv)

Fine-grained emotion recognition supports therapy tools and social robots, but it needs facial data, which raises privacy and data-protection concerns. EmoNet-Face-HQ answers that with generated portraits, expert-rated over a 40-category taxonomy far finer than the usual six to eight basic emotions. Under the protocol it ships with, vision-language models (VLMs) score poorly on that taxonomy, and the benchmark concludes that a dedicated fine-tuned model is necessary: Empathic-Insight-Face (EIF; Small/Large). We show that off-the-shelf VLMs match or beat that fine-tuned model when the answer is not generated but read from the logits, as one binary query per category. We keep the benchmark's images, taxonomy and ratings, and change only how the answer is read. Experts agree at kappaw = 0.468 on the five categories they measure most reliably. Generatively, no interval among eleven open-weight VLMs lies entirely above that anchor (kappaw=0.268-0.486). Under verification all eleven clear it, each of them significantly better at kappaw=0.507-0.586. Three also significantly beat EIF sitting at kappaw = 0.551 (Small; 0.534 Large). The gain comes from the graded probability and not from asking a yes/no question: as a control, thresholding those same probabilities to yes/no costs 142% of the average gains and drops binarization below generative elicitation to kappaw=0.254-0.423. A replication on real photographs (FACES) is weaker and mixed: of the ten models that pass a validity gate, six gain, three are neutral to positive and one is negative, so the effect is not confined to synthetic data.