# Disentangled Source-Free Personalization for Facial Expression Recognition with Neutral Target Data Source: [Disentangled Source-Free Personalization for Facial Expression Recognition with Neutral Target Data]() Collection index: [llms.txt]() Data updated: 2025-10-26T04:52:00.639Z Authors: Masoumeh Sharafi, Emma Ollivier, Muhammad Osama Zeeshan, Soufiane Belharbi, Marco Pedersoli, Alessandro Lameiras Koerich, Simon Bacon, Eric Granger Published: 2025-03-26T17:53:53.000Z ## Abstract Facial Expression Recognition (FER) from videos is a crucial task in various application areas, such as human-computer interaction and health diagnosis and monitoring (e.g., assessing pain and depression). Beyond the challenges of recognizing subtle emotional or health states, the effectiveness of deep FER models is often hindered by the considerable inter-subject variability in expressions. Source-free (unsupervised) domain adaptation (SFDA) methods may be employed to adapt a pre-trained source model using only unlabeled target domain data, thereby avoiding data privacy, storage, and transmission issues. Typically, SFDA methods adapt to a target domain dataset corresponding to an entire population and assume it includes data from all recognition classes. However, collecting such comprehensive target data can be difficult or even impossible for FER in healthcare applications. In many real-world scenarios, it may be feasible to collect a short neutral control video (which displays only neutral expressions) from target subjects before deployment. These videos can be used to adapt a model to better handle the variability of expressions among subjects. This paper introduces the Disentangled SFDA (DSFDA) method to address the challenge posed by adapting models with missing target expression data. DSFDA leverages data from a neutral target control video for end-to-end generation and adaptation of target data with missing non-neutral data. Our method learns to disentangle features related to expressions and identity while generating the missing non-neutral expression data for the target subject, thereby enhancing model accuracy. Additionally, our self-supervision strategy improves model adaptation by reconstructing target images that maintain the same identity and source expression. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Ambivalence/hesitancy recognition on BAH 10 target subjects, 214 source subjects](): Ambivalence/hesitancy recognition; BAH 10 target subjects, 214 source subjects; Subject 1 Performance: 61.24. - [Stress Recognition on StressID](): Stress Recognition; StressID; Sub-1 Score: 73.47. - [Facial Expression Recognition on BAH 214 source subjects (10 target subjects)](): Facial Expression Recognition; BAH 214 source subjects (10 target subjects); Accuracy (Sub-1): 49.1. - [Pain Recognition on BioVid 77 source subjects 10 target subjects](): Pain Recognition; BioVid 77 source subjects 10 target subjects; Subject 1 Accuracy: 77. - [Facial Expression Recognition on BioVid (target subjects (10))](): Facial Expression Recognition; BioVid (target subjects (10)); Accuracy (Sub-1): 65.72. - [Facial Expression Recognition on Aff-Wild2 (10 target subjects)](): Facial Expression Recognition; Aff-Wild2 (10 target subjects); Accuracy (Subject 1): 58.31. ## Original sources and implementation [Original paper]() [PDF]()