Search

The goal of gaze estimation is to estimate a gaze vector from an image containing a face or eye(s). Most existing studies use pre-defined fixed-resolution images to estimate the gaze vector. However, images captured from in-the-wild environments may have various resolutions, and variation in resolution can degrade gaze estimation performance. To address this problem, a gaze estimation method from arbitrary-sized low-resolution images is proposed. The basic idea of the proposed method is to combine knowledge distillation and feature adaptation. Knowledge distillation helps the gaze estimator for arbitrary-sized images generate a feature map similar to that from a high-resolution image. Feature adaptation makes creating a feature map adaptive to various resolutions of an input image possible by using a low-resolution image and its scale information together. It is shown that combining these two ideas improves gaze estimation performance substantially in the ablation study. It is also demonstrated that the proposed method can be generalized to other popularly used gaze estimation models through experiments using various backbones. [ABSTRACT FROM AUTHOR]

Three-dimensional human mesh reconstruction from a single video has made much progress in recent years due to the advances in deep learning. However, previous methods still often reconstruct temporally noisy pose and mesh sequences given in-the-wild video data. To address this problem, we propose a human pose refinement network (HPR-Net) based on a non-local attention mechanism. The pipeline of the proposed framework consists of a weight-regression module, a weighted-averaging module, and a skinned multi-person linear (SMPL) module. First, the weight-regression module creates pose affinity weights from a 3D human pose sequence represented in a unit quaternion form. Next, the weighted-averaging module generates a refined 3D pose sequence by performing temporal weighted averaging using the generated affinity weights. Finally, the refined pose sequence is converted into a human mesh sequence using the SMPL module. HPR-Net is a simple but effective post-processing network that can substantially improve the accuracy and temporal smoothness of 3D human mesh sequences obtained from an input video by existing human mesh reconstruction methods. Our experiments show that the noisy results of the existing methods are consistently improved using the proposed method on various real datasets. Notably, our proposed method reduces the pose and acceleration errors of VIBE, the existing state-of-the-art human mesh reconstruction method, by 1.4% and 66.5%, respectively, on the 3DPW dataset. [ABSTRACT FROM AUTHOR]

In this paper, we address the problem of 3D human mesh reconstruction from a single 2D human pose based on deep learning. We propose MeshLifter, a network that estimates a 3D human mesh from an input 2D human pose. Unlike most existing 3D human mesh reconstruction studies that train models using paired 2D and 3D data, we propose a weakly supervised learning method based on a loop structure to train the MeshLifter. The proposed method alleviates the difficulty of obtaining ground-truth 3D data to ensure that the MeshLifter can be trained successfully from a 2D human pose dataset and an unpaired 3D motion capture dataset. We compare the proposed method with recent state-of-the-art studies through various experiments and show that the proposed method achieves effective 3D human mesh reconstruction performance. Notably, our proposed method achieves a reconstruction error of 59.1 mm without using the 3D ground-truth data of Human3.6M, the standard dataset for 3D human mesh reconstruction. [ABSTRACT FROM AUTHOR]

Although the performance of the 3D human shape reconstruction method has improved considerably in recent years, most methods focus on a single person, reconstruct a root-relative 3D shape, and rely on ground-truth information about the absolute depth to convert the reconstruction result to the camera coordinate system. In this paper, we propose an end-to-end learning-based model for single-shot, 3D, multi-person shape reconstruction in the camera coordinate system from a single RGB image. Our network produces output tensors divided into grid cells to reconstruct the 3D shapes of multiple persons in a single-shot manner, where each grid cell contains information about the subject. Moreover, our network predicts the absolute position of the root joint while reconstructing the root-relative 3D shape, which enables reconstructing the 3D shapes of multiple persons in the camera coordinate system. The proposed network can be learned in an end-to-end manner and process images at about 37 fps to perform the 3D multi-person shape reconstruction task in real time. [ABSTRACT FROM AUTHOR]