Panoptic Segmentation aims to further the process of understanding an image by classifying all the pixels in it while also assigning them to unique object instances. I designed and implemented a top-down method by extending the existing MaskRCNN model with a semantic segmentation head to generate both instance and semantic segmentation masks. I also implemented a post-processing layer that combines the resulting masks.

Organization

Student

Srihari Humbarwadi

Mentors

  • jaeyounkim
  • Morgan Roff
  • Abdullah Rashwan
  • Margaret Maynard-Reid
  • Sayak Paul
close

2021