This is a release of FlowNet-S and FlowNet-C. It comes as a fork of the caffe master branch and with a trained network, as well as examples on how to use or train it.
To get started with FlowNet, first compile caffe, by configuring a
"Makefile.config" (example given in Makefile.config.example)
then make with
$ make -j 5 all
Go to this folder:
./flownet/models/flownet/
From this folder you can execute the scripts we prepared: To try out FlowNetS on a sample image pair, run
./demo_flownet.py S data/0000000-img0.ppm data/0000000-img1.ppm
You can also provide lists of files to run it on multiple image pairs. To train FlowNetS with the 8 sample images that come with this package, just run:
./train_flownet.py S
To extend it, please modify the img1_list.txt and img2_list.txt files accordingly or adapt the python script for your needs. Please use strong image augmentation techniques to obtain satisfactory results.
Please cite this paper in your publications if you use FlowNet for your research:
@inproceedings{DFIB15,
author = "A. Dosovitskiy and P. Fischer and E. Ilg and P. H{\"a}usser and C. Haz\ırba\ş and V. Golkov and P. v.d. Smagt and D. Cremers and T. Brox",
title = "FlowNet: Learning Optical Flow with Convolutional Networks",
booktitle = "IEEE International Conference on Computer Vision (ICCV)",
month = "Dec",
year = "2015",
url = "http://lmb.informatik.uni-freiburg.de/Publications/2015/DFIB15"
}