Generative adversarial framework for depth filling via Wasserstein metric, cosine transform and domain transfer
作者：
Highlights：
• Depth completion can be performed by learning the context and contents of the scene.
• Depth holes can be predicted based on the scene and capture device characteristics.
• Using absolute deviations loss in frequency domain (DCT) improves reconstruction.
• Adversarial training (Wasserstein metric) can improve training and mode selection.
• Using domain transfer, models trained on synthetic data are used in the real world.
摘要
•Depth completion can be performed by learning the context and contents of the scene.•Depth holes can be predicted based on the scene and capture device characteristics.•Using absolute deviations loss in frequency domain (DCT) improves reconstruction.•Adversarial training (Wasserstein metric) can improve training and mode selection.•Using domain transfer, models trained on synthetic data are used in the real world.
论文关键词：Depth image,Hole filling,Self-supervised learning,Generative model,Adversarial training,Feature distance,Domain adaptation
论文评审过程：Received 1 March 2018, Revised 18 December 2018, Accepted 17 February 2019, Available online 27 February 2019, Version of Record 4 March 2019.
论文官网地址：https://doi.org/10.1016/j.patcog.2019.02.010
