search

plemeri / InSPyReNet

Official PyTorch implementation of Revisiting Image Pyramid Structure for High Resolution Salient Object Detection (ACCV 2022)
MIT License
321 stars 61 forks source link

Empty mBA result for PASCAL-S dataset #22

Closed moktsuiqin closed 1 year ago

moktsuiqin commented 1 year ago

Hi, can I know what is the reason of being NaN for mBA result only in PASCAL-S dataset?

moktsuiqin commented 1 year ago

Screenshot 2023-03-24 074352

plemeri commented 1 year ago

First of all, bounday accuracy (mBA) is not designed for Low Resolution datasets (DUTS-TE, DUT-OMRON, ECSSD, HKU-IS, PASCAL-S). Second, unlike other datasets, PASCAL-S has multiple values other than 255 according to each object's saliency. In other words, while other datasets are labeled either 0 or 255 for each pixel, PASCAL-S has multiple values including 0 and 255 such as 128.

Long story short, because of such difference, mBA shows NaN. Thanks

moktsuiqin commented 1 year ago

Thank you so much for your answer