Open ChristophReich1996 opened 2 months ago
Dear authors,
Congratulations on the great work, and thank you for providing the code. Unfortunately, it is quite hard to use. I am also interested in the evaluation. Running the COCO evaluation as provided in the repo and README instructions is not possible, as parts of the code still seem to be missing. Furthermore, it would be great if you could provide the missing details for the Cityscapes evaluation, as it is impossible to compare to your method without having those missing information.
I would greatly appreciate your help! Thanks!
Best regards, Oliver
Hello, for the Cityscapes, for thing, we use 8 categories, 11 categories for stuff @ChristophReich1996
We should have test the evaluation code on COCO should work, but for cityscapes, we cannot retrieve the old script since our server is updated through the summer. Feel free to leave message here about the details of the missing file of running evaluation on COCO :) @olvrhhn
Hi @Dantong88, @frank-xwang & @XinyangHan,
We are currently using U2Seg as our baseline. To this end, we would like to match the numbers reported in the paper. However, both from the paper and from the code, it is not clear how many classes have been utilized for validation on Cityscapes. Do you use the 27 classes, the 19 classes, or the 7 parent classes? Thanks for your help!