This repository contains the code for applying datamosh or 'glitchy' visual effect in videos.

Example datamoshed videos with different video shots -- panning, zooming, looping, etc: YouTube Playlist

Setup

Clone the datamosh repository and then recursively clone the optical flow RAFT dependency.

git clone https://www.github.com/CCareaga/datamosh.git
git clone --recurse-submodules -j8 https://github.com/princeton-vl/RAFT.git

conda create -n datamosh
conda activate datamosh

Follow RAFT guidelines to install the dependencies

conda install pytorch=1.6.0 torchvision=0.7.0 cudatoolkit=10.1 matplotlib tensorboard scipy opencv -c pytorch

Additional python dependencies for datamosh

pip install kornia tqdm

Demo

1. Animating Looping

python loop_mosh.py --input_path ./examples/video_01.MOV --output_path ./examples/looping.gif --gif

---

2. Panning

For videos with panning or zooming shots, use the mosh.py with either a pair of videos or a directory of videos. Refer to the following command line arguments to run the script. Additionally, you can choose between different optical flow frameworks. The default framework is Farneback, to use RAFT, set --raft. The default datamosh effect is in forward direction, to create the reverse effect use --reverse.

optional arguments:
  -h, --help            show this help message and exit
  -p1 PATH_1, --path_1 PATH_1
                        path to first video
  -p2 PATH_2, --path_2 PATH_2
                        path to second video
  -if INPUT_FOLDER, --input_folder INPUT_FOLDER
                        folder of input videos (ordered by name)
  -op OUT_PATH, --out_path OUT_PATH
                        path to output video, e.g. ./output.mp4
  -rv, --reverse        flag to reverse the output
  -rt, --raft           flag to use raft flow
  -ri RAFT_ITER, --raft_iter RAFT_ITER
                        raft iterations
  -fs FLOW_SPEED, --flow_speed FLOW_SPEED
                        optical flow speed
  -fp FLOW_PERC, --flow_perc FLOW_PERC
                        threshold on magnitude before dropping pixels
  -fh HEIGHT, --height HEIGHT
                        frame height
  -fw WIDTH, --width WIDTH
                        frame height

python mosh.py --input_path ./<video_directory>/ --output_path ./<output.mp4>

3. Zooming

Credits

We have borrowed code from the following projects:

1. RAFT
2. Farneback