Optical flow dataset loader

README.md

@@ -13,3 +13,19 @@ If you use this code, please cite:
 https://gist.github.com/fvisin/7104500ae8b33c3b65798d5d2707ce6c#file-dataset_loaders-bib))
 
 Check the full documentation on: http://dataset_loaders.readthedocs.io
+### Optical flow
+The dataset loaders can optionally load from disk, or in some cases compute,
+the optical flow associated to the video sequences. To do so it looks for a
+file in `<dataset_path>/OF/<OF_type>/prefix/filename.npy>` where prefix is the
+name of the subset (or video) as returned by get_names(). If the file is
+missing it will try to compute the optical flow for the entire dataset once and
+store it on disk.
+
+At the moment the only optical flow algorithm supported to this end is the
+Farneback (requires openCV installed, choose 'Farn' as type), but you can
+easily pre-compute the optical flow with your preferred algorithm and then load
+it via the dataset loaders. An example code for a few algorithms is provided
+[here](https://github.com/marcociccone/opencv_of_gpu).
+NO SUPPORT WILL BE PROVIDED FOR THIS CODE OR ANY OTHER OPTICAL FLOW CODE NOT
+DIRECTLY INTEGRATED IN THIS FRAMEWORK.
+