search

sacmehta / ESPNet

ESPNet: Efficient Spatial Pyramid of Dilated Convolutions for Semantic Segmentation
https://sacmehta.github.io/ESPNet/
MIT License
541 stars 112 forks source link

testing on unseen dataset (Mapillary) #35

Closed shaoyuanlo closed 5 years ago

shaoyuanlo commented 5 years ago

Hi, In your paper, you conducted an experiment that trained ESPNet on Cityscapes but tested on Mapillary. However, the number of classes is different between the two datasets. How did you test and evaluate the mIoU? Thanks.

sacmehta commented 5 years ago

1) We tested on category level and not at class level. 2) classes belonging to the same category are merged (e.g.: car and truck are merged to vehicle category)

shaoyuanlo commented 5 years ago

How did you get the category level labels?

sacmehta commented 5 years ago

You follow the same procedure as Cityscapes does.

pkuqgg commented 5 years ago

How did you get the category level labels?

Hi!I am a student, do you have Mapillary data set? Have you solved the label problems? Hoping for your reply! Thanks!