search

B1ueber2y / TrianFlow

Code release for "Towards Better Generalization: Joint Depth-Pose Learning without PoseNet".
MIT License
267 stars 40 forks source link

How to predict depth aligned with pose? #16

Open ahangchen opened 4 years ago

ahangchen commented 4 years ago

I want to align the depth by your single image depth prediction branch with the pose predicted by the pose branch during inference. But it seems that you don't use pose when depth inference? I only find model.infer_depth() in the test code which only use single image. I'm afraid that depth consistency without pose aligned is not guaranteed.

liyingliu commented 4 years ago

@ahangchen I guess you could refer to the function process_video in infer_vo.py (https://github.com/B1ueber2y/TrianFlow/blob/master/infer_vo.py#L161)