GitHub - AIoT-Lab-BKAI/OT-4-Video-Fewshot

Branches Tags

Name		Name	Last commit message	Last commit date
Latest commit History 15 Commits
__pycache__		__pycache__
configs		configs
datasets		datasets
deep_speed		deep_speed
job_output		job_output
job_scripts		job_scripts
models		models
runs		runs
splits		splits
tools		tools
trainers		trainers
utils		utils
videotransforms		videotransforms
.gitignore		.gitignore
Readme		Readme
basic_module.py		basic_module.py
env.txt		env.txt
env.yaml		env.yaml
script.sh		script.sh
secrets.yaml		secrets.yaml
setup.py		setup.py
test.py		test.py
test.yaml		test.yaml

Repository files navigation

Download UCF Data using this command in terminal
kaggle datasets download -d pevogam/ucf101

After download data, process each video into a folder of frame, the folder has the same name as the video but without the suffix .avi
The data directory has the following structure
data
    ucf101
        ApplyEyeMakeup
            v_ApplyEyeMakeup_g01_c01
                00000.jpg
                00001.jpg
                00002.jpg
                00003.jpg
            v_ApplyEyeMakeup_g01_c02
            v_ApplyEyeMakeup_g01_c03


To setup conda enviroment run this command
conda env create -f env.yml

To run training script, use this bash command
PYTORCH_CUDA_ALLOC_CONF=max_split_size_mb:512  python runs/train_molo_ot.py