JakeCowton/pytorch-segmentation-starter

Semantic and instance segmentation using a custom dataset.

A generic Instance and Semantic Segmentation Starter Repo

Semantic and instance segmentation implementations that operate on VOC, COCO, & Data Science Bowl 2018 (implemented as a custom dataset). If you wish to modify this for your own dataset, use the corresponding dsb_dataset.py as a template.

This was developed in order to carry out instance and semantic segmentation for finding nucliei in micorscope images of blood cells in order to do blood cancer detection.

Supported datasets

• COCO
• Pascal VOC
• Data Science Bowl 2018 Nucleus detection in blood cells

Supported models

• FCN ResNet101
• DeepLabV3 ResNet101
• Mask R-CNN

Demo Scripts

See the respective README files in detection_ref and segmentation_ref.