In this study, we proposed a micro-expression recognition framework combining the Vision Transformer (ViT) network based on the multi-head self-attention mechanism with a pre-trained model. In this framework, each image in the micro-and-macro expression warehouse (MMEW) dataset was divided into 224 patches, and these patches were then rearranged into a one-dimensional vector. Due to the difficulty of collection, the datasets of micro-expressions are usually small, making it difficult to achieve satisfactory results using traditional deep network models. Therefore, in our framework we introduce a pre-trained model (ViT_base_patch16_224) that is trained on ImageNet to address this problem. Given that the pre-trained model has sufficient parameters and the MMEW dataset is smaller, a freezing strategy is adopted to make it suitable for the MMEW data. To the best of our knowledge, this is the first work that combining used the ViT model and the pre-trained model for micro-expression recognition task. Our preliminary experiments results show that the proposed micro-expression recognition framework achieved an accuracy of 87% on MMEW after 10 epochs training, which is the state-of-the-art accuracy achieved in this data set. Moreover, the accuracy is further improved to 93% with 30 epochs training, reflecting the potential of our framework in micro-expression recognition.


    Zugriff

    Zugriff prüfen

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Titel :

    Micro-expression Recognition Using Pre-trained Model and Transformer


    Beteiligte:
    Fan, Yali (Autor:in) / Jia, Minghan (Autor:in) / Zhang, Yifan (Autor:in) / Yu, Yang (Autor:in)


    Erscheinungsdatum :

    2022-10-12


    Format / Umfang :

    1389274 byte




    Medientyp :

    Aufsatz (Konferenz)


    Format :

    Elektronische Ressource


    Sprache :

    Englisch



    Underwater Image Enhancement Using Pre-trained Transformer

    Boudiaf, Abderrahmene / Guo, Yuhang / Ghimire, Adarsh et al. | British Library Conference Proceedings | 2022



    RAINDROP RECOGNITION DEVICE, VEHICULAR CONTROL APPARATUS, METHOD OF TRAINING MODEL, AND TRAINED MODEL

    BANNO YOSHIHISA / TAKANO SHO / INOGUCHI MASAAKI et al. | Europäisches Patentamt | 2021

    Freier Zugriff

    Raindrop recognition device, vehicular control apparatus, method of training model, and trained model

    BANNO YOSHIHISA / TAKANO SHO / INOGUCHI MASAAKI et al. | Europäisches Patentamt | 2023

    Freier Zugriff

    Micro-expression Recognition Using a Shallow ConvLSTM-Based Network

    Shukla, Saurav / Rai, Prabodh Kant / Verlekar, Tanmay T. | British Library Conference Proceedings | 2023