mirror of
https://github.com/qurator-spk/sbb_pixelwise_segmentation.git
synced 2025-06-09 11:50:04 +02:00
Update README
This commit is contained in:
parent
cd1990d4f9
commit
a216dccfcf
1 changed files with 8 additions and 5 deletions
13
README
13
README
|
@ -4,17 +4,20 @@ how to train:
|
|||
|
||||
format of ground truth:
|
||||
|
||||
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 from 0 , 1 ,2 .., n_classes-1.
|
||||
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.
|
||||
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.
|
||||
|
||||
traing , evaluation and output:
|
||||
training , evaluation and output:
|
||||
train and evaluation folder should have subfolder of images and labels.
|
||||
And output folder should be free folder which the output model will be written there.
|
||||
And output folder should be empty folder which the output model will be written there.
|
||||
|
||||
patches:
|
||||
|
||||
|
|
Loading…
Add table
Add a link
Reference in a new issue