Make sure you have the correct file permission and environment
conda env create -f environment.yml # creates a conda env with the name "acmai"
conda activate acmai # activate conda environment
chmod a+x ./tools/vid2frame.sh # give executing permission to the bash script
To download and unzip SumMe dataset
python ./tools/fetch_dataset.py
To convert videos into .jpg frames in identically titled subfolders
./tools/vid2frame.sh
To read and convert labels into annotation.txt
to facilitate data loading into pytorch
python ./tools/readmat.py
If you want to use colab or other platforms that require you to upload files on to a remote server the following allows you to convert the .jpg frames from each video into single frames.hdf5
file.
python ./tools/hdf5.py
Note: our dataloader is NOT yet compatible with this file format
TODO:
- implement loss history plotting
- clean up `annotation.txt (possibly make it relative to each video for easier coding)
- keep
requirements.txt
/environment.yml
up to date - collect all the scripts into a util folder / unified script
- look into a different file format of faster dataset upload to gdrive (hdf5?)
- write scripts to standardize video resolution (downsample) (explore options: opencv? pytorch? ffmpeg?)
- make sure the scripts work cross-platform (currently macos is ok)
- fix label association of the data image