search

lliuz / ARFlow

The official PyTorch implementation of the paper "Learning by Analogy: Reliable Supervision from Transformations for Unsupervised Optical Flow Estimation".
MIT License
251 stars 50 forks source link

utils.flow_utils ==> flow_to image #16

Closed RokiaAbdeen closed 3 years ago

RokiaAbdeen commented 3 years ago

the output flow of kitti dataset is with only one color , what is the wrong? 33

lliuz commented 3 years ago

Can you provide more details, such as the image names and some code snippets, so that I can reproduce your problem?

lliuz commented 3 years ago

I'm not sure if the cause of the problem is the same as #15, if you are using pytorch >= 1.3.0, please pull the latest code, and tell me if the problem still exists.

RokiaAbdeen commented 3 years ago

k1

RokiaAbdeen commented 3 years ago

k2

RokiaAbdeen commented 3 years ago

here is it, the complete inference code, with torch==1.1.0

lliuz commented 3 years ago

You used images from the left and right view at the same time, _10 in the filename means the first frame for a pair of frames, image_2 means left view, image_3 means right view in KITTI dataset. So all pixels are just moving to left with blue in visualization.

RokiaAbdeen commented 3 years ago

oh I understand, so to select two consecutive frames , I should use both of them from one folder , image_2 or image_3, right?

lliuz commented 3 years ago

yes

RokiaAbdeen commented 3 years ago

thanks so much for your answers