Implementation of PseudoEdgeNet: Nuclei Segmentation only with Point Annotations

dataset:

• https://nucleisegmentationbenchmark.weebly.com/

Paper:

• PseudoEdgeNet: Nuclei Segmentation only with Point Annotations

Implementation and Result:

• training time: around 30 sec on a single NVIDIA Tesla V100 with 16 GB memory
• batch_size = 2
• epochs = 60
• could produce same result showing the paper if use same evalutaion method (10-fold CV)

structure

• Note.ipynb: how I implement it, thoughts, methods
• main.ipynb: main function file
• libs: source code

dependency:

• tensorflow >= 2.2.0
• imgaug == 0.4.0

Usage:

• Download the dataset using the link.
• create a folder named 'dataset' (no quotes), under the dataset folder, create two folder named 'Annotations' and 'Tissue images', separately (no quotas). Then put *.xml under 'Annotations' folder, put *.png under 'Tissue images' (from the dataset you download).
• put main.ipynb, libs folder, dataset folder in same directory.
• run main.ipynb cell by cell...