flow

The GIF is pretty large, might take a minute to load here on Github. Give it a few. You can also check out the video in the images folder (clip.mp4)

Intro

Was just scrolling through PyTorch docs and stumbled upon optical flow, figured I should try it out.

Video used is from blender see acknowledgements below

Getting started

Dependencies

To set up your python environment to run the code in this repository, follow the instructions below.

1. Create (and activate) a new environment with Python 3.10.6.

   • Linux or Mac:

   conda create --name flow python=3.10.6
   source activate flow

   • Windows:

   conda create --name flow python=3.10.6
   activate flow

2. Clone the repository (if you haven't already!), and navigate to the flow/ folder. Then, install several dependencies.

git clone https://github.com/mrdvince/flow.git
cd flow
pip install -r requirements.txt

3. Run the thing Incase your run out of memory try setting to a smaller number,

i.e e.g to=30 generates up to the 30th second of the clip

python flow.py --video <video source> --to <time in seconds you want the video to stop at>

Do note, running this on cuda will be much faster and using the M1 gpu doesn't work currently due to missing ops.

Acknowledgements

• Coffe run blender video
• Optical flow post on pytorch docs