anuragranj / humanflow2

Official repository of Learning Multi-Human Optical Flow (IJCV 2019)
Other
38 stars 3 forks source link

demo for run_inference.py #1

Open feitongt opened 4 years ago

feitongt commented 4 years ago

Hi,

Could you provide a demo code for inferencing the optical flow of a testing image pair.

I am trying to run the run_inference.py, but the result looks not good. I am not sure if I run the code in the right way, thus I hope to have an official demo code to test image pairs.

Thanks, Feitong

anuragranj commented 4 years ago

You should use the test_humanflow.py and wrap your dataset around it. Depending on the model, you would need to preprocess differently: SPyNet - Normalize (default) and use --div-flow 1 PWC-Net - Prevent normalization using --no-norm and use --div-flow 20