Dual CNN Models for Unsupervised Monocular Depth Estimation
share
A lot of progress has been made to solve the depth estimation problem in stereo vision. Though, a very satisfactory performance is observed by utilizing the deep learning in supervised manner for depth estimation. This approach needs huge amount of ground truth training data as well as depth maps which is very laborious to prepare and many times it is not available in real scenario. Thus, the unsupervised depth estimation is the recent trend by utilizing the binocular stereo images to get rid of depth map ground truth. In unsupervised depth computation, the disparity images are generated by training the CNN with an image reconstruction loss based on the epipolar geometry constraints. The effective way of using CNN as well as investigating the better losses for the said problem needs to be addressed. In this paper, a dual CNN based model is presented for unsupervised depth estimation with 6 losses (DNM6) with individual CNN for each view to generate the corresponding disparity map. The proposed dual CNN model is also extended with 12 losses (DNM12) by utilizing the cross disparities. The presented DNM6 and DNM12 models are tested over KITTI driving database and compared with the recent state-of-the-art result of unsupervised depth estimation.
READ FULL TEXT