reproducing the results on inbreast dataset

[seyedaminhabibi]

Hi, thanks for your great code. i tried to reproduce your result on inBreast dataset nut i couldn't get any similar or even close results. i tried to preprocess inbreast dicom images as you said in yor paper. is there any specific preprocessing that we should do? could you please add some code so we can regenerate your result on inbreast?

[riblidezso]

Hi,

Sorry for the late answer, been busy lately.

There is no other preprocessing than described in this section of the paper:

The images had low contrast, therefore we adjusted the window of the pixel levels. The pixel values were clipped to be minimum 500 pixel lower and maximum 800 pixels higher than the mode of the pixel value distribution (excluding the background) and were rescaled to the 0–255 range.

The mode of the pixel values was determined for every single image.

Otherwise the demo notebook should produce the the same results as in the paper.
Note, that you should use the published prototxt files, and the the published weights.

You also need to use the bounding box annotations published here on github to get quantitative results.
https://github.com/riblidezso/frcnn_cad/blob/master/data/inbreast_ground_truth_cancer_bbox_rois.tsv

Good luck for your work!