Mainstream methods treat head pose estimation as a supervised classification/regression problem, whose performance heavily depends on the accuracy of ground-truth labels of training data. However, it is rather difficult to obtain accurate head pose labels in practice, due to the lack of effective equipment and reasonable approaches for head pose labeling. In this paper, we propose a method which does not need to be trained with head pose labels, but matches the keypoints between a reconstructed 3D face model and the 2D input image, for head pose estimation. The proposed head pose estimation method consists of two components: the 3D face reconstruction and the 3D–2D matching keypoints. At the 3D face reconstruction phase, a personalized 3D face model is reconstructed from the input head image using convolutional neural networks, which are jointly optimized by an asymmetric Euclidean loss and a keypoint loss. At the 3D–2D keypoints matching phase, an iterative optimization algorithm is proposed to match the keypoints between the reconstructed 3D face model and the 2D input image efficiently under the constraint of perspective transformation. The proposed method is extensively evaluated on five widely used head pose estimation datasets, including Pointing’04, BIWI, AFLW2000, Multi-PIE, and Pandora. The experimental results demonstrate that the proposed method achieves excellent cross-dataset performance and surpasses most of the existing state-of-the-art approaches, with average MAEs of ${4.78}^{\circ }$ on Pointing’04, ${6.83}^{\circ }$ on BIWI, ${7.05}^{\circ }$ on AFLW2000, ${5.47}^{\circ }$ on Multi-PIE, and ${5.06}^{\circ }$ on Pandora, although the model of the proposed method is not trained on any of these five datasets.

Facial expression recognition (FER) is a challenging problem due to the intra-class variation caused by subject identities. In this paper, a self-difference convolutional network (SD-CNN) is proposed to address the intra-class variation issue in FER. First, the SD-CNN uses a conditional generative adversarial network to generate the six typical facial expressions for the same subject in the testing image. Second, six compact and light-weighted difference-based CNNs, called DiffNets, are designed for classifying facial expressions. Each DiffNet extracts a pair of deep features from the testing image and one of the six synthesized expression images, and compares the difference between the deep feature pair. In this way, any potential facial expression in the testing image has an opportunity to be compared with the synthesized “Self”—an image of the same subject with the same facial expression as the testing image. As most of the self-difference features of the images with the same facial expression gather tightly in the feature space, the intra-class variation issue is significantly alleviated. The proposed SD-CNN is extensively evaluated on two widely-used facial expression datasets: CK+ and Oulu-CASIA. Experimental results demonstrate that the SD-CNN achieves state-of-the-art performance with accuracies of 99.7% on CK+ and 91.3% on Oulu-CASIA, respectively. Moreover, the model size of the online processing part of the SD-CNN is only 9.54 MB (1.59 MB $\times 6$), which enables the SD-CNN to run on low-cost hardware.

Various applications require realistic, artifact-free, and animatable 3D avatars. However, traditional 3D morphable models (3DMMs) produce animatable 3D heads but fail to capture accurate geometries and details, while existing deep implicit functions have been shown to achieve realistic reconstructions but suffer from artifacts and struggle to yield 3D heads that are easy to animate. To reconstruct high-fidelity, artifact-less, and animatable 3D heads from single-view images, we leverage semantics to bridge the best properties of 3DMMs and deep implicit functions and propose SeIF—a semantic-constrained deep implicit function. First, SeIF derives fine-grained semantics from a standard 3DMM (e.g., FLAME) and samples a semantic code for each query point in the query space to provide a soft constraint to the deep implicit function. The reconstruction results show that this semantic constraint does not weaken the powerful representation ability of the deep implicit function while significantly suppressing artifacts. Second, SeIF predicts a more accurate semantic code for each query point and utilizes the semantic codes to uniformize the structure of reconstructed 3D head meshes with the standard 3DMM. Since our reconstructed 3D head meshes have the same structure as the 3DMM, 3DMM-based animation approaches can be easily transferred to animate our reconstructed 3D heads. As a result, SeIF can reconstruct high-fidelity, artifact-less, and animatable 3D heads from single-view images of individuals with diverse ages, genders, races, and facial expressions. Quantitative and qualitative experimental results on seven datasets show that SeIF outperforms existing state-of-the-art methods by a large margin. The code and models of SeIF are available for research purposes at https://github.com/starVisionTeam/SeIF .

Background: Facial behavior has emerged as a crucial biomarker for autism identification. However, heterogeneity among individuals with autism poses a significant obstacle to traditional feature extraction methods, which often lack the necessary discriminative power. While deep-learning methods hold promise, they are often criticized for their lack of interpretability. Methods: To address these challenges, we developed an innovative facial behavior characterization model that integrates coarse- and fine-grained analyses for intelligent autism identification. The coarse-grained analysis provides a holistic view by computing statistical measures related to facial behavior characteristics. In contrast, the fine-grained component uncovers subtle temporal fluctuations by employing a long short-term memory (LSTM) model to capture the temporal dynamics of head pose, facial expression intensity, and expression types. To fully harness the strengths of both analyses, we implemented a feature-level attention mechanism. This not only enhances the model’s interpretability but also provides valuable insights by highlighting the most influential features through attention weights. Results: Upon evaluation using three-fold cross-validation on a self-constructed autism dataset, our integrated approach achieved an average recognition accuracy of 88.74%, surpassing the standalone coarse-grained analysis by 8.49%. Conclusions: This experimental result underscores the improved generalizability of facial behavior features and effectively mitigates the complexities stemming from the pronounced intragroup variability of those with autism, thereby contributing to more accurate and interpretable autism identification.