/pyflow

Fast, accurate and easy to run dense optical flow with python wrapper

Primary LanguageC++OtherNOASSERTION

Python Dense Optical Flow

Python wrapper for Ce Liu's C++ implementation of Coarse2Fine Optical Flow. This is super fast and accurate optical flow method based on Coarse2Fine warping method from Thomas Brox. This python wrapper has minimal dependencies, and it also eliminates the need for C++ OpenCV library. For real time performance, one can additionally resize the images to a smaller size.

Run the following steps to download, install and demo the library:

git clone https://github.com/pathak22/pyflow.git
cd pyflow/
python setup.py build_ext -i
python demo.py    # -viz option to visualize output

Process video

Process a video and save the extracted of images and frames into two folders. Command: python proc_video.py --vid_src [video_src] --of_dir [of_folder_path] --img_dir [frames_folder_path]

The algorithm credits to CVPR 2017 paper on Unsupervised Learning using unlabeled videos.