computer vision - How to calculate the mean IU score in image segmentation? -
how compute mean iu (mean intersection on union) score in this paper?
long, jonathan, evan shelhamer, , trevor darrell. "fully convolutional networks semantic segmentation."
for each class intersection on union (iu) score is:
true positive / (true positive + false positive + false negative)
the mean iu average on classes.
regarding notation in paper:
- n_cl : number of classes
- t_i : total number of pixels in class i
n_ij : number of pixels of class i predicted belong class j. class i:
- n_ii : number of correctly classified pixels (true positives)
- n_ij : number of pixels wrongly classified (false positives)
- n_ji : number of pixels wrongly not classifed (false negatives)
you can find matlab code compute directly in pascak devkit here
Comments
Post a Comment