Project to run video style transfer on Edge TP
This has been tested in python=3.9.X. Please create a virtual environment and install the requirements:
Windows
py -m venv edge
.\edge\Scripts\activate
pip install -r requirements.txt
Mac/Ubuntu
python3 -m venv edge
source edge/bin/activate
pip install -r requirements.txt
All critical run defaults for training and inference can be found in our argument parser in
reconet_tensorflow/utils/parser.py.
Full training can be run using:
python3 reconet_tensorflow/train.py
During development, you can run the debug mode by passing the flag
python3 reconet_tensorflow/train.py --debug True
You can run the style inference via your webcam on your local machine (note that you might need to give permission on your machine) using the default arguments. Note that the default aspect ratios are set to that of MacBook webcams
python3 reconet_tensorflow/video_demo.py
python3 reconet_tensorflow/video_demo.py --video-input sample_vid.mp4 --video-output styled_video.mp4
You can also specify the input/output resolution using (here 512x216 for input, 1080x760 for output)
python3 reconet_tensorflow/video_demo.py --input-resolution 512, 360 --output-resolution 1080, 760
