MEGC2026: Micro-Expression Grand Challenge on Visual Question Answering

Xinqi Fan; Jingting Li; John See; Moi Hoon Yap; Su-Jing Wang; Adrian K. Davison

cs.CVMar 9, 2026

MEGC2026: Micro-Expression Grand Challenge on Visual Question Answering

Authors: Xinqi Fan, Jingting Li, John See, Moi Hoon Yap, Su-Jing Wang, Adrian K. Davison

Organizations: Department of Computing and Mathematics, Manchester Metropolitan University · State Key Laboratory of Cognitive Science and Mental Health, Institute of Psychology, CAS & Department of Psychology, University of the Chinese Academy of Sciences · School of Mathematical and Computer Sciences, Heriot-Watt University Malaysia

Abstract

Facial micro-expressions (MEs) are involuntary movements of the face that occur spontaneously when a person experiences an emotion but attempts to suppress or repress the facial expression, typically found in a high-stakes environment. In recent years, substantial advancements have been made in the areas of ME recognition, spotting, and generation. The emergence of multimodal large language models (MLLMs) and large vision-language models (LVLMs) offers promising new avenues for enhancing ME analysis through their powerful multimodal reasoning capabilities. The ME grand challenge (MEGC) 2026 introduces two tasks that reflect these evolving research directions: (1) ME video question answering (ME-VQA), which explores ME understanding through visual question answering on relatively short video sequences, leveraging MLLMs or LVLMs to address diverse question types related to MEs; and (2) ME long-video question answering (ME-LVQA), which extends VQA to long-duration video sequences in realistic settings, requiring models to handle temporal reasoning and subtle micro-expression detection across extended time periods. All participating algorithms are required to submit their results on a public leaderboard. More details are available at https://megc2026.github.io.

Figures & tables

Fig. 1: An overview of Micro-Expression Grand Challenges (MEGCs).

Method	SAMM						CAS(ME) 3						Overall
	Coarse		Fine-Grained				Coarse		Fine-Grained				Coarse		Fine-Grained
	UF1	UAR	UF1	UAR	BLEU	ROUGE	UF1	UAR	UF1	UAR	BLEU	ROUGE	UF1	UAR	UF1	UAR	BLEU	ROUGE
Qwen2.5VL-3B (ZS)	0.242	0.333	0	0	0.090	0.295	0.262	0.333	0.057	0.048	0.089	0.315	0.256	0.333	0.057	0.048	0.065	0.312
Qwen2.5VL-3B (FT)	0.206	0.25	0.057	0.1	0.234	0.331	0.376	0.515	0.044	0.111	0.109	0.375	0.328	0.422	0.059	0.117	0.122	0.362
Qwen3VL-4B (ZS)	0.242	0.333	0	0	0.068	0.229	0.262	0.382	0.114	0.222	0.065	0.382	0.256	0.333	0.079	0.200	0.066	0.339
Qwen3VL-4B (FT)	0.242	0.333	0.178	0.250	0.128	0.416	0.262	0.333	0	0	0.236	0.413	0.256	0.333	0.052	0.064	0.204	0.415

TABLE I: Baseline results of the ME-VQA task with ME-VQA-v2 training and test sets. The test set results of the ME-VQA task using Qwen2.5VL-3B and Qwen3VL-4B (the baseline methods) trained on the curated ME-VQA dataset. The coarse emotion class refers to positive, negative, and surprise. The fine-grained emotion class refers to happiness, surprise, fear, disgust, anger, and sadness. ZS refers to zero-shot performance, while FT refers to fine-tuned performance.

	Expression		ME		MaE		AU		Expression Type
Method	MAE	RMSE	MAE	RMSE	MAE	RMSE	F1	Jaccard	UF1	UAR
Qwen2.5VL-3B (ZS)	5.80	6.48	3.10	3.89	2.10	2.70	0.161	0.098	0.373	0.490
Qwen2.5VL-3B (FT)	5.20	6.05	3.20	4.02	2.10	2.70	0.169	0.110	0.560	0.636
Qwen3VL-4B (ZS)	6.20	6.84	3.60	4.27	3.10	3.54	0.075	0.048	0.479	0.525
Qwen3VL-4B (FT)	5.50	6.35	3.30	4.11	3.00	3.44	0.167	0.103	0.487	0.591

TABLE II: Baseline results of the ME-LVQA task. The test set results of the ME-LVQA task using Qwen2.5VL-3B and Qwen3VL-4B (the baseline methods) trained on the curated ME-LVQA dataset. The fine-tuning used 5 subjects from SAMM-LV and 5 subjects from CAS(ME) 3 . ZS refers to zero-shot performance, while FT refers to fine-tuned performance.

Micro-Expression Recognition

See all 12 papers

MagPlus: Bridging Micro-to-Regular Facial Expressions through Learnable Magnification
Jun 11, 2026Sliman Jammal, Andrei SharfFacial Expression Synthesis
Ben-Gurion University of the Negev, Israel
MA-Bench: Towards Fine-grained Micro-Action Understanding
Mar 27, 2026Kun Li, Jihao Gu, Fei Wang +3Multimodal Large Language Models Multimodal Model Evaluation
CVLab, College of Information Technology, United Arab Emirates University · University College London · Hefei University of Technology +2
When, Where, and Why: Hierarchical Causal Explainability for Video Face Understanding with MLLMs
Oct 13, 2026Anubhav Kataria, Shreya Ghosh, Jingwen Ye +1VLM Interpretability Facial Expression Recognition
Monash University, Melbourne, Australia · Indian Institute of Technology Kharagpur, Kharagpur, India

Jun 11, 2026J/K move · Enter open · S save

MagPlus: Bridging Micro-to-Regular Facial Expressions through Learnable Magnification

Sliman Jammal, Andrei Sharf

Ben-Gurion University of the Negev, Israel

Facial Expression Synthesis

Open paper AskPDF arXiv

Facial micro-expressions are subtle and short-lived facial movements that provide important cues about genuine human emotions. However, modeling and generating them remains difficult because annotated micro-expression data is limited and the underlying facial motions are extremely weak. Existing micro-expression generation methods therefore often suffer from limited quality, weak robustness, and poor generalization. We propose MagPlus, a transferable micro-expression processing pipeline that connects micro-expression analysis with standard facial animation models. Instead of training a dedicated generator from scratch, MagPlus learns to magnify subtle facial motions into the range of regular facial expressions, transforming micro-expressions into signals that are compatible with existing facial expression processing models. The magnified sequence is then used by a standard facial expression model for tasks such as transfer and synthesis. A complementary DeMagPlus module then restores the generated motion back to realistic micro-expression intensity levels while preserving the synthesized dynamics. We evaluate the framework using four facial animation models: FOMM, FSRT, MetaPortrait, and EmoPortraits. None of these models are trained on micro-expression data. Experiments show that MagPlus-DeMagPlus enables pretrained macro-expression models to generate more realistic micro-expression motion without retraining the backbones.

Abstract

Micro-Expression Recognition

Explore similar work

MagPlus: Bridging Micro-to-Regular Facial Expressions through Learnable Magnification

MA-Bench: Towards Fine-grained Micro-Action Understanding

When, Where, and Why: Hierarchical Causal Explainability for Video Face Understanding with MLLMs