/cell-segm-tracker

Stem cells segmentation and tracking

Primary LanguageJupyter Notebook

Stem cells segmentation and tracking on videos via Detectron2

There are images (frames) with mesenchymal stem cells (MSC). There is a need to track individual cells and analyse the dinamics of their movements on videos (cell migration analysis).

Pipeline

  • instance segmentation with Mask R-CNN R-50/101 FPN
  • tracking with centroid_multi_tracker

Labeled dataset

  • train: 137 images (2010 cells)
  • test: 57 images (722 cells)

Tools and libs

  • python 3.7
  • Labelme 3.16 (labeling)
  • Clodsa 1.2 (augmentation)
  • OpenCV 3.4
  • Pytorch 1.5
  • Detectron2 0.1.3

Training on Google Colaboratory (GPU, Tesla P100)

Metrics

Good predictions and errors