1

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?

pedro_galher
  • 334
  • 8
  • 18

1 Answers1

1

This is a pretty old question, but I got here thorough Google so why not.

To train Mask-RCNN, you need to create a separate binary mask for each instance. Many image segmentation tools do that automatically. I personally used LabelBox in my most recent project, but there are many open source, free software as well. Here is a list I found on Google, but I cannot really recommend anything specific.

liorr
  • 764
  • 5
  • 21