Testing issues with accuracies (global, class avg, mean I/O)

[jinuhjoo]

Hello All,

I have a problem calculating accuracies on my test result image data files. In the mean time global acc are looking good around 85% but the rest seems to be poor under 15%

This is what I've done : Used test_segmentation_camvid.py to save my resulting prediction_rgb and groundtruth_rgb (e.g scipy.misc.toimage(rgb, cmin=0.0, cmax=255).save(image_name) scipy.misc.toimage(rgb_gt, cmin=0.0, cmax=255).save(image_name))

And used matlab script provided below https://github.com/alexgkendall/SegNet-Tutorial/blob/master/Scripts/compute_test_results.m

Training data are provided CamVid data and trained through SegNet and SegNet-Basic

Could there be something wrong with the scripts or have I missed something? Please help.

P.S. if there is a python version of compute_test_results.m could you kindly upload it. (not really good at python or matlab)

[mrnaif799]

Hi @jinuhjoo,

I am facing the same situation, did you find solve the problem?

Thanks,

[Alice-kenan]

@jinuhjoo @mrnaif799 I am facing the same situation, did you find solve the problem?

Thanks