This repo contains the PyTorch version of MatchNet.
- scikit-video: accessing images / videos
pip install sk-video
- tqdm: progress bar
conda config --add channels conda-forge
conda update --all
conda install tqdm
utils/resize_and_split.sh
: Prepares your (video) data for training. Resizes videos present in folders of folders (i.e. directory of classes) and may split them into training and validation set. May also skip short videos and trim longer ones.main.py
: Script to start training. Use-h
to print the command line arguments help.
python -u main.py <CLI arguments> | tee train.log
To run on a specific GPU, say n
, type CUDA_VISIBLE_DEVICES=n
just before python ...
.