Ycb_Yolov5_Trainer

Installation

Commands to be executed in order:

1. git clone Ycb_Yolov5_Trainer (Clone to any folder other than your Ycb_Dataset_Generator folder)
2. cd ./Ycb_Yolov5_Trainer
3. Setup and install Python 3.7.12 or Python 3.8.0 using Pyenv. CREATE and ACTIVATE a new python environment named 'YCB_Trainer' running Python 3.7.12 or Python 3.8.0. (This step is required to successfully meet all the requirements)
   • pyenv virtualenv 3.7.12 YCB_Trainer (Install python and create python virtual environment)
   • (Only for the initial setup) pyenv local YCB_Trainer (Activate python virtual environment)
   • pyenv versions (Check if the correct virtual environment with intended python version is active)
4. pip install -r ./Requirements/requirements.txt (Installs Ycb_Yolov5_Trainer's requirements)
5. Move/copy your custom dataset generated from following the steps on Ycb_Dataset_Generator to the current directory.

   •  mv ../Ycb_Dataset_Generator/custom_dataset .

   OR

   •  cp ../Ycb_Dataset_Generator/custom_dataset .

6. git clone yolov5 (Clones ultralytics's yolov5 repository)
7. cd ./yolov5
8. pip install -r requirements.txt (Installs ultralytics's yolov5's requirements)
9. Training a model (Transfer learning with ultralytic's pretrained checkpoints - faster & accurate):

   •  python ./train.py --img 640 --batch 16 --epochs 500 --data ../custom_dataset/data.yaml --cfg ./models/custom_yolov5s6.yaml --weights 'yolov5s6.pt' --name yolov5s6_results --cache

   • pretrained weights (more checkpoints available on ultralytics/yolov5 repository): yolov5s6
   • epochs (more number of epochs possible): 500
   • batch (higher number is possible with better hardware)
10. Deploying custom trained model (Real Time Detection):
    • Collab's implementation:

      python ../yolov5Deploy.py live -1 ./runs/train/yolov5s6_results/weights/best.pt --mode 1

    • Ultralytic's implementation:

      python ./detect.py --source -1 --weights ./runs/train/yolov5s6_results/weights/best.pt