search

kinredon / Harmonious-Teacher

Official implementation of the CVPR 2023 paper "Harmonious Teacher for Cross-domain Object Detection"
Apache License 2.0
27 stars 3 forks source link

Cityscapes Imagesets #18

Open Thomas-Kb opened 1 day ago

Thomas-Kb commented 1 day ago

Dear Author, I have a question regarding the data for cityscapes -> foggy-cityscapes experiment: Which ImageSets did you use for the source and the target domain? Did you use all 2,975 images in both the source and the target domain or did you split it? If so, how did you split it? Thanks and best regards

kinredon commented 1 day ago

@Thomas-Kb Following previous works, we train the model using the labeled train split of cityscapes (2975 images) and the unlabeled train split of foggy-cityscapes (2975 images). The model is evaluated on the validation set (500 images) of foggy-cityscapes.