This is the data and code behind training the ML object detection model for findingwaldo.app
- python3 (preferred 3.10.x or newer)
pip install -r requirements.txt
- get training data:
wget https://storage.googleapis.com/public-file-server/waldo_training_images.tar
wget https://storage.googleapis.com/public-file-server/waldo_validation_images.tar
- untar images into the train and validate folders respectively
python train.py
- use a larger pretrained model as your base
- sizes: nano, small, medium, large, extra large
- trade off is speed vs accuracy
- outputs to a new 'run' directory with models and graphs/metrics of each of your training runs
python detect.py
- repo includes the nano sized model and test image
- trained in Vertex AI Workbench with n1-standard-4 machine type with an NVIDIA Tesla T4 GPU
