the large stride

[LIUTIANKUO]

Hi, thank you for your great idea!
When I reproduce your work in infer_vo.py, the function def get_prediction() #129 shows that it needs depth-net to predict depth1 and depth2, thus we can get the scale to align_to_depth in process_video #141.
But the code does not load the pretrained depth model when infer vo, right?
I think that is why then I sample the odometry dataset with different stride, my result is really bad.
So how can I solve the problem to get a good result with stride 2 or 3 ??