Collecting such annotated datasets is challenging. We propose to complete existing databases by generating new database entries.
We test our algorithm on four challenging benchmark datasets (ICVL, MSRA, NYU and Big Hand 2.2M datasets) and demonstrate that our approach outperforms or is on par with state-of-the-art methods quantitatively and qualitatively
한줄요약
3D 생성은 처음이지, 3D 손 데이터 생성기, GAN기반의 Discriminator를 포즈추적으로
Institute: Imperial College London URL: https://arxiv.org/pdf/1805.04497.pdf Keyword: HandPoseEstimator, HandPoseGenerator (GAN) Interest: 2
Summary
한줄요약