minhhungvu17/lesions_segmentation

retinal-lesion-torch

Pytorch Implementation of Retinal Lesion Detection and Segmentation

Using detectron2

Prerequisites

1. CUDA 10.1

2. pycocotools: Installation

3. Install packages*:

pip install -r requirments.txt

*Compatible with CUDA 10.1

Data Structures

.
+-- images
|   +-- <CTEH-xxxxx.jpg>
|   +-- ...
+-- lesions
|   +-- hemorrhage
|   +-- <CTEH-xxxxx.jpg>
|   +-- ...
|   +-- exudates
|   +-- <CTEH-xxxxx.jpg>
|   +-- ...
|   +--  microaneurysms
|   +--  <CTEH-xxxxx.jpg>
|   +--  ...
+-- instances_default.json

Procedure

1. Run data_preprocessing to generate train_set.json and val_set.json

   Script:

   python data_preprocessing --data path/to/data/folder --annotation path/to/annotation/file --output path/to/output/folder
   

   Default:

   python data_preprocessing --data ./data --annotation ./data/instances_default.json --output ./data