文件名称：cvpr18-3D Human Pose Estimation in the Wild by Adversarial Learning
文件大小：1.89MB
文件格式：PDF
更新时间：2021-07-24 07:36:07
3D人体姿态 GAN DNN
Recently, remarkable advances have been achieved in 3D human pose estimation from monocular images because of the powerful Deep Convolutional Neural Networks (DC- NNs). Despite their success on large-scale datasets col- lected in the constrained lab environment, it is difficult to obtain the 3D pose annotations for in-the-wild images. Therefore, 3D human pose estimation in the wild is still a challenge. In this paper, we propose an adversarial learn- ing framework, which distills the 3D human pose structures learned from the fully annotated dataset to in-the-wild im- ages with only 2D pose annotations. Instead of defining hard-coded rules to constrain the pose estimation results, we design a novel multi-source discriminator to distinguish the predicted 3D poses from the ground-truth, which helps to enforce the pose estimator to generate anthropometri- cally valid poses even with images in the wild. We also observe that a carefully designed information source for the discriminator is essential to boost the performance. Thus, we design a geometric descriptor, which computes the pair- wise relative locations and distances between body joints, as a new information source for the discriminator. The ef- ficacy of our adversarial learning framework with the new geometric descriptor has been demonstrated through exten- sive experiments on widely used public benchmarks. Our approach significantly improves the performance compared with previous state-of-the-art approaches.