Large language models (LLMs) have recently been applied to time series forecasting to leverage their reasoning and pattern recognition capabilities. Compared to task-specific forecasting models, LLMs exhibit generalizability and a broad understanding of cross-domain knowledge. However, current LLM-based forecasting methods overlook the importance of frequency properties in sequence data, which is a critical aspect in time series analysis. In this work, we propose an approach to incorporate frequency domain representation and operations into an LLM-based forecasting framework. We transform the label sequences into Fourier complex-valued representations and adapt LLMs to forecast in the frequency domain. To enhance frequency analysis and prediction, a Fourier neural network is introduced in the LLM-based forecasting. Extensive experiments verify that our approach compares favorably against the state-of-the-art methods in time series forecasting.

Estimating optical flow from facial videos is an essential preprocessing step for many applications. However, it is a challenging task as the facial videos contain rich expressions, large displacements, and complex occlusions. Obtaining the ground truth optical flow for facial videos is very difficult, which hinders the supervised learning of optical flow from monocular in-the-wild facial videos. In this paper, we provide an effective and accurate method for the unsupervised learning of optical flow from facial videos. An occlusion-aware global-local matching model is introduced for the joint reasoning of optical flow and occlusions. We propose a novel occlusion estimation paradigm to detect occlusions caused by facial expressions and pose variations. Experiments demonstrate that our method compares favorably against the state-of-the-art methods in facial optical flow estimation.

Automated segmentation of masticatory muscles is a challenging task considering ambiguous soft tissue attachments and image artifacts of low-radiation cone-beam computed tomography (CBCT) images. In this paper, we propose a bi-graph reasoning model (BGR) for the simultaneous detection and segmentation of multi-category masticatory muscles from CBCTs. The BGR exploits the local and long-range interdependencies of regions of interest and category-specific prior knowledge of masticatory muscles by reasoning on the category graph and the region graph. The category graph of the learnable muscle prior knowledge handles high-level dependencies of muscle categories, enhancing the feature representation with noise-agnostic category knowledge. The region graph models both local and global dependencies of the candidate muscle regions of interest. The proposed BGR accommodates the high-level dependencies and enhances the region features in the presence of entangled soft tissue and image artifacts. We evaluated the proposed approach by segmenting masticatory muscles on clinically acquired CBCTs. Extensive experimental results show that the BGR effectively segments masticatory muscles with state-of-the-art accuracy.

In this paper, we propose a fully convolutional network-based dense map from voxels to invertible pair of displacement vector fields regarding a template grid for the consistent voxel-wise correspondence. We parameterize the volumetric mapping using a convolutional network and train it in an unsupervised way by leveraging the spatial transformer to minimize the gap between the warped volumetric image and the template grid. Instead of learning the unidirectional map, we learn the nonlinear mapping functions for both forward and backward transformations. We introduce the combinational inverse constraints for the volumetric one-to-one maps, where the pairwise and triple constraints are utilized to learn the cycle-consistent correspondence maps between volumes. Experiments on both synthetic and clinically captured volumetric cone-beam CT (CBCT) images show that the proposed framework is effective and competitive against state-of-the-art deformable registration techniques.