search

NVIDIA / flownet2-pytorch

Pytorch implementation of FlowNet 2.0: Evolution of Optical Flow Estimation with Deep Networks
Other
3.08k stars 738 forks source link

Train and validate my own dataset #256

Open sotomotocross opened 3 years ago

sotomotocross commented 3 years ago

Hello everyone. I would like to generate my own optical flow dataset (i have specific video footage) and the trainig and validate FlowNet 2. Can anyone guide me both through the MPI Sintel dataset flow code to generate my dataset from my video footage and the training and testing of the present NN? Thank you very much in advance for the help. I have set up a docker image with CUDA 10 and ubuntu 18.04 and also pytorch 1.0.0 the NN has successfully complete the install.sh process and can run, but the documentation for some things doesn't make sense. I can't run the inference example. Also for the training i have to download all the FlyingChairs dataset?

Li-hunter commented 1 year ago

Training your own dataset? Do you have flow labels of the dataset?