[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-86062-en":3,"doc-seo-86062-105":28,"detail-sidebar-cat-0-en-105":89},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":4,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":11,"language":21,"language_code":22,"site_id":23,"html_lang":22,"table_of_contents":24,"faqs":25,"seo_title":13,"seo_description":14,"update_tm":26,"read_time":27},86062,687197207057,"Sage","https://ap-avatar.wpscdn.com/davatar_29158cc5080c5b710cf443261637dec0",8,"Research & Report","AU-Guided Synthetic Video Generation for Micro-Expression Recognition","Micro-expression recognition is constrained by small dataset scale, limited demographic coverage, and restricted emotion labels. EquiME is introduced as a synthetic micro-expression dataset built with an AU-guided image-to-video generation pipeline. It provides 75K videos from 15K source face images across five emotions, with automatically inferred demographic metadata and video-quality measurements. Evaluations use similarity, spatial variation, and no-reference perceptual-quality metrics plus cross-dataset MER experiments on SAMM and CASME II, showing competitive transfer and low variation across architectures.","AU-Guided Synthetic Video Generation for Micro-Expression  \nRecognition  \nPei-Sze Tan Sailaja Rajanala Yee-Fan Tan Rapha¨el C.-W. Phan  \nHuey-Fang Ong  \nCyPhi AI Lab, Monash University Malaysia  \narXiv :2607 . 10860v1 [ cs .CV] 12 Jul 2026  \nAbstract  \nMicro-expression recognition is limited by the small scale, narrow demographic coverage, and restricted emotion labels of existing datasets. We introduce EquiME, a synthetic micro-expression dataset built from AU-guided image-to-video generation. EquiME contains 75K videos generated from 15K source face images across five target emotions, together with automatically inferred demographic metadata and video-quality measurements. We evaluate EquiME using frame-pair similarity, spatial variation, and no-reference perceptual-quality metrics, together with cross-dataset MER experiments on SAMM and CASME II. Models trained on EquiME achieve competitive cross-dataset performance on SAMM and CASME II and show comparatively low variation across the four evaluated architectures. This paper focuses on the dataset design, the structured AU-conditioning pipeline used for video generation, and the empirical evidence needed to assess EquiME as a synthetic MER resource. Project page: [https://kirito](https://kirito)[blade.github.io/me-vlm/](blade.github.io/me-vlm/)  \nKeywords: Micro-expression, image-to-video model, facial action units, dataset generation  \n1 Introduction  \nMicro-expression recognition (MER) studies subtle and short-lived facial movements that can reveal underlying affective states. The problem is important for affective computing, human-computer interaction, and behavioral analysis, but progress remains constrained by the datasets available for training and evaluation [13, 21] . Compared with mainstream facial analysis tasks, MER still relies on small corpora collected in controlled environments, often with limited subject diversity and uneven emotion coverage [3, 11, 12, 15, 20, 23, 24] . These limitations make it difficult to train robust models and to assess whether performance generalizes across identities and demographic groups.  \nThe data bottleneck is not only about scale. Existing benchmark datasets are also narrow in demo-  \ngraphic coverage, which can increase the risk that MER systems overfit to dataset-specific appearance cues rather than learn expression dynamics that transfer across populations [4, 22] . This concern is especially relevant in MER because the signal of interest is weak: subtle muscle activations can easily be confounded by identity, lighting, recording conditions, and annotation noise. Standard evaluation settings such as leave-one-subject-out validation partly address subject overlap, but they do not fully resolve broader questions of generalization across demographic groups [1, 6] .  \nSynthetic data offers a practical way to expand MER training resources, but current synthetic alternatives remain limited. Prior work has explored image-based synthesis or AU-driven generation [14, 25–27], yet available datasets often tradeoff realism, controllability, temporal continuity, or demographic coverage. In particular, existing open synthetic resources do not provide the combination of temporally coherent micro-expression videos, source-face diversity, and rich metadata needed for rigorous cross-dataset evaluation.  \nOverview of EquiME. We present EquiME, a synthetic micro-expression video dataset designed for cross-dataset MER research. EquiME is generated from source face images using an AU-guided image-to-video pipeline. The key idea is to treat facial Action Units (AUs) as an intermediate control representation: they provide a compact way to specify subtle muscle activations while preserving an explicit mapping between facial movement descriptions and target emotion labels. Using this formulation, we generate short video clips that cover the temporal progression of an expression while maintaining source-face consistency across samples.  \nEquiME contain","cbCaifBsf1DYPDl2","https://ap.wps.com/l/cbCaifBsf1DYPDl2","pdf",1794457,1,"English","en",105,"# Introduction\n## Problem and data bottleneck\n## Synthetic data and limitations\n## Overview of EquiME\n# Related Work\n## Micro-Expression Dataset","[{\"question\":\"What problem does the paper address in micro-expression recognition datasets?\",\"answer\":\"It addresses limitations of existing datasets, including small scale, narrow demographic coverage, and restricted emotion label sets that hinder robust training and reliable generalization assessment.\"},{\"question\":\"How is EquiME generated, and what role do Action Units play?\",\"answer\":\"EquiME is generated from source face images using an AU-guided image-to-video pipeline. Facial Action Units act as an intermediate control representation to specify subtle muscle activations while linking facial motion descriptions to target emotion labels.\"},{\"question\":\"What evaluation methods are used to assess EquiME’s usefulness for MER?\",\"answer\":\"The paper evaluates using frame-pair similarity, spatial variation, and no-reference perceptual-quality metrics, and it also performs cross-dataset MER experiments on SAMM and CASME II to measure transfer performance.\"}]",1784208190,20,{"code":4,"msg":29,"data":30},"ok",{"site_id":23,"language":22,"slug":31,"title":13,"keywords":32,"description":14,"schema_data":33,"social_meta":84,"head_meta":86,"extra_data":88,"updated_unix":26},"au-guided-synthetic-video-generation-for-micro-expression-recognition","",{"@graph":34,"@context":83},[35,52,66],{"@type":36,"itemListElement":37},"BreadcrumbList",[38,42,46,49],{"item":39,"name":40,"@type":41,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":43,"name":44,"@type":41,"position":45},"https://docshare.wps.com/document/","Document",2,{"item":47,"name":12,"@type":41,"position":48},"https://docshare.wps.com/document/research-report/",3,{"item":50,"name":13,"@type":41,"position":51},"https://docshare.wps.com/document/au-guided-synthetic-video-generation-for-micro-expression-recognition/86062/",4,{"url":50,"name":13,"@type":53,"author":54,"headline":13,"publisher":56,"fileFormat":59,"inLanguage":22,"description":14,"dateModified":60,"datePublished":60,"encodingFormat":59,"isAccessibleForFree":61,"interactionStatistic":62},"DigitalDocument",{"name":9,"@type":55},"Person",{"url":39,"name":57,"@type":58},"DocShare","Organization","application/pdf","2026-07-16",true,{"@type":63,"interactionType":64,"userInteractionCount":4},"InteractionCounter",{"@type":65},"ViewAction",{"@type":67,"mainEntity":68},"FAQPage",[69,75,79],{"name":70,"@type":71,"acceptedAnswer":72},"What problem does the paper address in micro-expression recognition datasets?","Question",{"text":73,"@type":74},"It addresses limitations of existing datasets, including small scale, narrow demographic coverage, and restricted emotion label sets that hinder robust training and reliable generalization assessment.","Answer",{"name":76,"@type":71,"acceptedAnswer":77},"How is EquiME generated, and what role do Action Units play?",{"text":78,"@type":74},"EquiME is generated from source face images using an AU-guided image-to-video pipeline. Facial Action Units act as an intermediate control representation to specify subtle muscle activations while linking facial motion descriptions to target emotion labels.",{"name":80,"@type":71,"acceptedAnswer":81},"What evaluation methods are used to assess EquiME’s usefulness for MER?",{"text":82,"@type":74},"The paper evaluates using frame-pair similarity, spatial variation, and no-reference perceptual-quality metrics, and it also performs cross-dataset MER experiments on SAMM and CASME II to measure transfer performance.","https://schema.org",{"og:url":50,"og:type":85,"og:title":13,"og:site_name":57,"og:description":14},"article",{"robots":87,"canonical":50},"index,follow",{"doc_id":7,"site_id":23},{"code":4,"msg":5,"data":90},[91,95,99,103,108,113,118,121,125,128,132],{"id":20,"doc_module":4,"doc_module_name":44,"category_name":92,"show_sort_weight":93,"slug":94},"Story & Novel",90,"story-novel",{"id":45,"doc_module":4,"doc_module_name":44,"category_name":96,"show_sort_weight":97,"slug":98},"Literature",80,"literature",{"id":51,"doc_module":4,"doc_module_name":44,"category_name":100,"show_sort_weight":101,"slug":102},"Exam",70,"exam",{"id":104,"doc_module":4,"doc_module_name":44,"category_name":105,"show_sort_weight":106,"slug":107},5,"Comic",60,"comic",{"id":109,"doc_module":4,"doc_module_name":44,"category_name":110,"show_sort_weight":111,"slug":112},6,"Technology",50,"technology",{"id":114,"doc_module":4,"doc_module_name":44,"category_name":115,"show_sort_weight":116,"slug":117},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":44,"category_name":12,"show_sort_weight":119,"slug":120},30,"research-report",{"id":122,"doc_module":4,"doc_module_name":44,"category_name":123,"show_sort_weight":27,"slug":124},9,"Religion & Spirituality","religion-spirituality",{"id":27,"doc_module":4,"doc_module_name":44,"category_name":126,"show_sort_weight":27,"slug":127},"World Cup","world-cup",{"id":129,"doc_module":4,"doc_module_name":44,"category_name":130,"show_sort_weight":129,"slug":131},10,"Lifestyle","lifestyle",{"id":133,"doc_module":4,"doc_module_name":44,"category_name":134,"show_sort_weight":104,"slug":135},19,"General","general"]