From f69d4459cbd542db3b4a613283b255aa75f9070f Mon Sep 17 00:00:00 2001 From: "Rezanezhad, Vahid" Date: Thu, 5 Dec 2019 16:14:00 +0100 Subject: [PATCH] Add new file --- README.md | 26 ++++++++++++++++++++++++++ 1 file changed, 26 insertions(+) create mode 100644 README.md diff --git a/README.md b/README.md new file mode 100644 index 0000000..c4dc27e --- /dev/null +++ b/README.md @@ -0,0 +1,26 @@ +# 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. \ No newline at end of file