From bbe6f99a85fa8cb164442d3d685897cc2a1612b0 Mon Sep 17 00:00:00 2001 From: "Rezanezhad, Vahid" Date: Thu, 5 Dec 2019 16:13:37 +0100 Subject: [PATCH] Update README --- README | 27 +-------------------------- 1 file changed, 1 insertion(+), 26 deletions(-) diff --git a/README b/README index e103b0b..5237d53 100644 --- a/README +++ b/README @@ -1,27 +1,2 @@ -# Train - just run: python train.py with config_params.json - - -# Ground truth format - - Lables for each pixel is identified by a number . So if you have a binary case n_classes should be set to 2 and - labels should be 0 and 1 for each class and pixel. - In the case of multiclass just set n_classes to the number of classes you have and the try to produce the labels - by pixels set from 0 , 1 ,2 .., n_classes-1. - The labels format should be png. - - If you have an image label for binary case it should look like this: - - Label: [ [[1 0 0 1], [1 0 0 1] ,[1 0 0 1]], [[1 0 0 1], [1 0 0 1] ,[1 0 0 1]] ,[[1 0 0 1], [1 0 0 1] ,[1 0 0 1]] ] - this means that you have an image by 3*4*3 and pixel[0,0] belongs to class 1 and pixel[0,1] to class 0. - -# Training , evaluation and output - train and evaluation folder should have subfolder of images and labels. - And output folder should be empty folder which the output model will be written there. - -# Patches - - if you want to train your model with patches, the height and width of patches should be defined and also number of - batchs (how many patches should be seen by model by each iteration). - In the case that model should see the image once, like page extraction, the patches should be set to false. +