lyndonzheng / Synthetic2Realistic

[ECCV 2018]: T2Net: Synthetic-to-Realistic Translation for Depth Estimation Tasks
180 stars 42 forks source link

Questions about Kitti and VKitti datasets #41

Closed hiroyasuakada closed 3 years ago

hiroyasuakada commented 3 years ago

Thank you for your great work! I have two questions about Kitti and VKitti datasets.

1. About Kitti dataset In training, were only 22600 left images used? or both left and right images (45200 in total) used?

2. About Virtual Kitti dataset From some comments in Issue 5, it seems that you used 21030 images for training and 230 images for testing out of total 21260 VKitti images. (trainA_SYN.txt, testA_SYN.txt)

However, you mentioned in issue 20 that you got rid of rain, fog images for training in the ECCV 2018 paper.

Could you please tell me what exactly did you use in VKitti for training to get the result of the ECCV2018 paper?