liruoteng/FlowNet

prediction is flipped version of ground truth

Mehuli-Ruh11 opened this issue · 2 comments

Mehuli-Ruh11 commented

While visualizing the results, got by test_flownet.py on ChairSDHom dataset, the prediction is flipped 180 degree.
Ground truth:
figure_5
Prediction:
figure_1-5
Does anyone have idea about what could be the reason for this ?

liruoteng commented

I think it is because you use data layer to read the image and the layer has function to mirror the input image.

thanks
RT

Mehuli-Ruh11 commented

Yes ! That was the problem while reading the .flo file. Thanks !!