proper dataset format to use for object detection with retinanet code example from site keras.io

Question

im trying to use object detection with retinanet code example from keras.io on my own dataset, but i cant find out the proper data format. the code itself is trained on some