What is your input image to generate the .flo file with PWC-NET?

[byeciao]

I use pytorch-pwc to generate the .flo file but it seems to be no data within the object,
Firstly, the input to pwc I used is the original image,

after I generate the .flo file, I test the vdo_slam, but it seems no data in flow


then I convert it to .png file, but it seems to be no flow in the object:

[byeciao]

After I comment these lines, it only work on first 2 frames:

and it can't detect the moving car or estimate its speed.

[halajun]

Hi, @real-chenzy. Sorry for the late reply. I think the mask of object has been filtered in the tracking process, due to the object being too small in the scene. You can change the pre-set variable of "ThDepthOBJ" in the yaml file, or modify the thresholds in the function of Tracking::DynObjTracking(). Hope this may help.