search

meng-tang / rloss

Regularized Losses (rloss) for Weakly-supervised CNN Segmentation
MIT License
207 stars 47 forks source link

annotation label #16

Closed teeinn closed 4 years ago

teeinn commented 4 years ago

Hi, I am trying to implement training with pascal voc dataset in pytorch. While searching for the code that imports scribble annotation xml file, I realized it only imports images from JPEG folder (original image) and pascal 2012 scribble folder (scribble images, all scribble is black color). I might be misunderstanding your paper or code,, but how come training is working not knowing class label for each of lines all colored black? Waiting for your quick response!

meng-tang commented 4 years ago

The scribbles look black, but the pixel values are actually 0-20 which correspond to class labels

teeinn commented 4 years ago

oh!!!!!!!!!! I see!!! Thank you so much! it really does have different values! so then, are these images (pascal 2012 scribble dir) created based on xml annotation file? I mean, I know this dataset is from scribblesub (they mentioned amazon mechanical turk,) but do you have any transition code for the process?

meng-tang commented 4 years ago

yes, you can recreate from https://github.com/meng-tang/rloss/blob/master/data/pascal_scribble/convertscribbles.m (credit to http://www.jifengdai.org/downloads/scribble_sup )

teeinn commented 4 years ago

Thank you for your quick response!!