I want to train a MaskRCNN model using the Object detection API Tensorflow.
I'm currently annotating the images using the Pixel Annotation Tool.
My question is: Should I create a training data set with ONE segmented feature per image, or can I annotate several features (of the same class) in a sigle image?