Training

In case you are interested to train your own segmentation model.

1. Data

The data folder should be organised in the following structure

data
-- train
	-- imgs
		-- img1.jpg
	-- labels
		-- img1.jpg
-- val
	-- imgs
		-- img2.jpg
	-- labels
		-- imgs.jpg

If you have data organised in other formats, it might be convenient to adjust the SegDataset class.

A label image should be a binary image where positive pixels take the value 255 and negative pixels 0.

2. Visualisation

We use visdom to visualise the training progression. To start a visdom server

python -m visdom.server

Author

Korsuk Sirinukunwattana ([email protected])