1.基于左右一致性的无监督单目深度估计
Unsupervised monocular depth estimation with left-right consistency
loss由三部分组成:
(外观匹配损失)
(视差平滑度损失)
(左右一致性损失)
2.具有自我提升能力的立体匹配的自监督学习(找不到代码)
Self-Supervised Learning for Stereo Matching with Self-Improving Ability
3.基于二进制分类的立体深度估计
Bi3D: Stereo Depth Estimation via Binary Classifications