This repository was archived by the owner on Jul 2, 2021. It is now read-only.
possible bug in the way that mIoU is computed #950
Description
I notice that the results for miou does not match with miou that I manually compute.
here is an example, lets say pres and labels are two lists including the predictions and gt data
I can compute the confusion matrix via chainercv.evaluations.calc_semantic_segmentation_confusion I also can compute the miou via chainercv.evaluations.eval_semantic_segmentation(preds, labels)
the miou based on confusion matrix can be computed as np.nanmean(np.diag(confusion) / (confusion.sum(axis=1) + confusion.sum(axis=0) - np.diag(confusion))) and these results dont match with np.nanmean(chainercv.evaluations.eval_semantic_segmentation(preds, labels)['iou'])