lior1990 / instance-seg

instance segmentation with deep metric learning and context

Instance Segmentation via Deep Embeddings

We present new approaches for the Instance Segmentation problem that are based on "Semantic Instance Segmentation with a Discriminative Loss Function" (https://arxiv.org/pdf/1708.02551.pdf) framework. We propose different approaches such as: modifying the suggested loss function to take into consideration critical parts of the object (edges, center), using MRF as post-processing action in order to improve the segments' borders, and building a neural-network that is capable of segmenting a unique object within a group of similar objects.

For full documentation please refer to: