README

This is the data and code behind training the ML object detection model for findingwaldo.app

How-to

Requirements

• python3 (preferred 3.10.x or newer)
• pip install -r requirements.txt

Training

• 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

Detection

python detect.py

• repo includes the nano sized model and test image

Stats/metrics of the current deployed model (XLarge)

• trained in Vertex AI Workbench with n1-standard-4 machine type with an NVIDIA Tesla T4 GPU