I used Tensorflow Object Detection API for a custom dataset based on the instructions at this help document.As required , collected the dataset,annotated it in PASCAL VOC XML format,split into training and test sets,generated tfrecords. Pets configuration for custom object detection and localization is used.

I just want to proceed further and get segmentation boundary rather than bounding box. The link is clear till this point on directory structure and mechanism to prepare input data.

Instance Segmentation help is unclear to me on two aspects: -Directory structure -Is it required to supply segment boundaries equal in number to that of images in training and test sets ?

Notice that tfrecords are created using this script. Can this be reused to include segmentation data.

  • $\begingroup$ Matterport Mask RCNN Keras/Tensorflow library github.com/matterport/Mask_RCNN is a good starting point for instance segmentation. There are some good example notebooks and some datasets so maybe it might help. $\endgroup$
    – Jack Vial
    Apr 5 '18 at 13:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.