Model Training

Processing scripts are in the mrobosub_ml repo

• Use VideoToImages.py to turn GoPro videos into images
  • usage: python3 VideoToImages.py PATHTOVIDEOFILE [wb]
• Use Preprocessing.py to resize image to desired size
• Use LabelBox to generate a JSON file for training
• Use LabelAndPartition.py to turn JSON file into image (png files, moves original image files) and label (.txt files) path
  • usage: python3 generateTrainingFiles.py [JSONFILE]

Training runs on HPC Cluster, scripts are in mrobosub_ml/slurm-scripts

• Run setup.sh
  • "USAGE: ./setup.sh [training data git url] [test data git url]"
  • Clones yolo repo
  • Clones dataset repo
  • Setup Venv for training
• Edit Parameters.txt for training params
• Run ./sbatch.sh & to submit job
• More details in the README