Facial emotions are expressed through a combination of facial muscle movements, namely, the Facial Action Units (FAUs). FAU intensity estimation aims to estimate the intensity of a set of structurally dependent FAUs. Contrary to the existing works that focus on improving FAU intensity estimation performance, this study investigates how knowledge distillation (KD) incorporated into a training model can improve FAU intensity estimation efficiency while achieving the comparable level of performance. Given the intrinsic structural characteristics of FAU, it is desirable to distill deep structural relationships, namely, DSR-FAU, using heatmap regression. Our methodology is as follows: First, a feature map-level distillation loss is applied to ensure that the student network and the teacher network share similar feature distributions. Second, the region-wise and channel-wise relationship distillation loss functions are introduced to penalize the difference in structural relationships. Specifically, the region-wise relationship can be represented by the structural correlations across the facial features, whereas the channel-wise relationship is represented by the implicit FAU co-occurrence dependencies. Third, we compare the model performance of DSR-FAU with the state-of-the-art models, based on two benchmarking datasets. It is shown that our model achieves comparable performance, with a lower number of model parameters and lower computation complexities.

Automatic gesture synthesis from speech is a topic that has attracted researchers for applications in remote communication, video games and Metaverse. Learning the mapping between speech and 3D full-body gestures is difficult due to the stochastic nature of the problem and the lack of a rich cross-modal dataset that is needed for training. In this paper, we propose a novel transformer-based framework for automatic 3D body gesture synthesis from speech. To learn the stochastic nature of the body gesture during speech, we propose a variational transformer to effectively model a probabilistic distribution over gestures, which can produce diverse gestures during inference. Furthermore, we introduce a mode positional embedding layer to capture the different motion speeds in different speaking modes. To cope with the scarcity of data, we design an intra-modal pre-training scheme that can learn the complex mapping between the speech and the 3D gesture from a limited amount of data. Our system is trained with either the Trinity speech-gesture dataset or the Talking With Hands 16.2M dataset. The results show that our system can produce more realistic, appropriate, and diverse body gestures compared to existing state-of-the-art approaches.

Speech-driven 3D facial animation is challenging due to the complex geometry of human faces and the limited availability of 3D audio-visual data. Prior works typically focus on learning phoneme-level features of short audio windows with limited context, occasionally resulting in inaccurate lip movements. To tackle this limitation, we propose a Transformer-based autoregressive model, Face-Former, which encodes the long-term audio context and autoregressively predicts a sequence of animated 3D face meshes. To cope with the data scarcity issue, we integrate the self-supervised pre-trained speech representations. Also, we devise two biased attention mechanisms well suited to this specific task, including the biased cross-modal multi-head (MH) attention and the biased causal MH self-attention with a periodic positional encoding strategy. The former effectively aligns the audio-motion modalities, whereas the latter offers abilities to generalize to longer audio sequences. Extensive experiments and a perceptual user study show that our approach outperforms the existing state-of-the-arts. The code and the video are available at: https://evelynfan.github.io/audio2face/

Facial expression recognition (FER) is crucial for social communication. However, current studies present limitations when addressing facial expression difference due to demographic variation, such as race, gender, and age, etc. In this article, we first propose a deeply-supervised attention network (DSAN) to recognize human emotions based on facial images automatically. Based on DSAN, a two-stage training scheme is designed, taking full advantage of the race/gender/age-related information. In our DSAN framework, multi-scale features are leveraged to capture more discriminative information from the deep layers to the shallow layers. Furthermore, we adopt the attention block to highlight the essential local facial characteristics; it performs well when it is incorporated into the deeply-supervised framework. Finally, we combine the complementary characteristics of multiple convolutional layers in deeply-supervised manner and ensemble the intermediate predicted scores. Our experimental results have shown that our proposed framework can (i) effectively integrate demographic information in improving the performance of a variety of FER tasks, (ii) learn informative feature representations with a visual explanation by capturing the regions of interests (ROI), (iii) achieve superior performance for both the posed and the spontaneous FER databases, each containing pictures of human facial expressions varied in gender, age or race.

The intensity estimation of facial action units (AUs) is challenging due to subtle changes in the person's facial appearance. Previous approaches mainly rely on probabilistic models or predefined rules for modeling co-occurrence relationships among AUs, leading to limited generalization. In contrast, we present a new learning framework that automatically learns the latent relationships of AUs via establishing semantic correspondences between feature maps. In the heatmap regression-based network, feature maps preserve rich semantic information associated with AU intensities and locations. Moreover, the AU co-occurring pattern can be reflected by activating a set of feature channels, where each channel encodes a specific visual pattern of AU. This motivates us to model the correlation among feature channels, which implicitly represents the co-occurrence relationship of AU intensity levels. Specifically, we introduce a semantic correspondence convolution (SCC) module to dynamically compute the correspondences from deep and low resolution feature maps, and thus enhancing the discriminability of features. The experimental results demonstrate the effectiveness and the superior performance of our method on two benchmark datasets.

Facial action unit (AU) intensity estimation aims to measure the intensity of different facial muscle movements. The external knowledge such as AU co-occurrence relationship is typically leveraged to improve performance. However, the AU characteristics may vary among individuals due to different physiological structures of human faces. To this end, we propose a novel geometry-guided representation learning (G2RL) method for facial AU intensity estimation. Specifically, our backbone model is based on a heatmap regression framework, where the produced heatmaps reflect rich information associated with AU intensities and their spatial distributions. Besides, we incorporate the external geometric knowledge into the backbone model to guide the training process via a learned projection matrix. The experimental results on two benchmark datasets demonstrate that our method is comparable with the state-of-the-art approaches, and validate the effectiveness of incorporating external geometric knowledge for facial AU intensity estimation.

Cross-dataset facial emotion recognition (FER) aims to reduce the discrepancy between the source and the target facial database. The topic is very challenging in FER, where facial features differ across different domains, such as ethnicity, age, gender and environmental condition. In practice, the labels of target facial expression database may be unavailable, making it impossible to fine-tune a pre-trained model via supervised transfer learning. To address this issue, we propose an unsupervised domain adaptation framework with adversarial learning for cross-dataset FER. We perform cross-dataset FER on three well-known publicly available facial expression databases, viz. CK+, Oulu-CASIA, and RAF-DB, showcasing the efficiency of our proposed approach.

Emotion recognition (ER) based on natural facial images/videos has been studied for some years and considered a comparatively hot topic in the field of affective computing. However, it remains a challenge to perform ER in the wild, given the noises generated from head pose, face deformation, and illumination variation. To address this challenge, motivated by recent progress in Convolutional Neural Network (CNN), we develop a novel deeply supervised CNN (DSN) architecture, taking the multi-level and multi-scale features extracted from different convolutional layers to provide a more advanced representation of ER. By embedding a series of side-output layers, our DSN model provides class-wise supervision and integrates predictions from multiple layers. Finally, our team ranked 3rd at the EmotiW 2018 challenge with our model achieving an accuracy of 61.1%.