Object detection trainer

finetune object detection models on your own data

More models and dataset formats will be added in the future but for now these are available

Supported models:

• fasterrcnn_resnet50_fpn

Supported dataset formats :

• MS COCO JSON

FIRST AND FOREMOST ANNOTATE A DATASET OR GET A DATASET TO WORK WITH AND PLACE IT IN THE 'data' DIRECTORY THEN IT SHOULD LOOK LIKE THIS

data/
├── train/
├── test/
└── valid/

installation

This was made with python 3.11 if you're on another version and it doesn't work consider using python 3.11

• create a python virtual env and activate it

python -m venv .venv
cd .venv
cd Scripts
activate
cd ..
cd ..

• install pytorch with cuda 12.4

 pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu124

• install other requirements.txt

pip install -r requirements.txt

Running the scripts

• you can use the default hyper parameters or open config.yaml and change them

• To train, log metrics, and test a model

python main.py

• To only test a trained model

cd src
python tester.py --model-name=your_model_name_here
# model names are timestamps which will be a number that looks like 1742073877

Now check the tests directory and you'll find it filled with comparisons between the model's prediction and the ground truth

• view logs in tensorboard

tensorboard --logdir=logs/