RetinaNet Player and Ball Detection

This project shows you how you can build a Computer Vision model that can detect players and ball in overhead camera images using RetinaNet.

This project uses Fizyr's awesome keras-retinanet implementation, and applies it to Haizaha's Soccer Player and Ball Detection free dataset.

Setup

You will need the Tensorflow and Keras installed in addition to the standard shenanigan.

The Data

To get the data and the pre-trained weight, run download.sh:

bash download.sh

This will download everything you need to run the project, including downloading and compiling keras-retinanet

Inference

To test the pre-trained model on a sample image, just run predict.py

python predict.py

Training

To train the model:

python keras-retinanet/keras_retinanet/bin/train.py --snapshot pre-trained/resnet50_csv_last.h5 csv data/train.csv data/labels.csv --val-annotations data/valid.csv

Before you can use the newly trained model, you'll need to convert it into an inference model:

python keras-retinanet/keras_retinanet/bin/convert_model.py ./snapshots/resnet50_csv_01.h5 ./snapshots/resnet50_csv_01_inference.h5