How's the pre-trained semantic segmentation model trained on, Cityscape or KITTI?

[ruili3]

Hi,

Thanks for your solid work and the datasets!

As you provided the pre-computed semantic labels for KITTI using method of Zhu etal , I wonder if the semantic segmentation model you use is trianed on KITTI or Cityscape? Or a mixture of KITTI and Cityscape? I notice there is no pretrained models and training code for KITTI dataset on Zhu's github page. Thanks a lot! :D

[TWJianNuo]

According to their paper, it is pretrained on Cityscape and finetuned with kitti 200 semantics segmentation label.

[ruili3]

thank you a lot！

TWJianNuo notifications@github.com 于2020年11月11日周三 下午9:58写道：

According to their paper http://openaccess.thecvf.com/content_CVPR_2019/papers/Zhu_Improving_Semantic_Segmentation_via_Video_Propagation_and_Label_Relaxation_CVPR_2019_paper.pdf, it is pretrained on Cityscape and finetuned with kitti 200 semantics segmentation label.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub https://github.com/TWJianNuo/EdgeDepth-Release/issues/7#issuecomment-725438224, or unsubscribe https://github.com/notifications/unsubscribe-auth/AFL646Y4A7LL235E4F3B7I3SPKKCBANCNFSM4TR7W6AQ .

[TWJianNuo]

Oh, It seems he removes pretrained model on kitti, strange. I used the model provided by the author.(but I do not know whether I still keep that.) In my memory, their model for kitti is pretrained using Cityscape's data plus Vista's data, then select hyper parameter for kitti under a 10-fold cross validation split of the 200 semantics data. Finally, they do a finetuning using kitti 200 semantics data with selected hyperparameter. By the way, tehre is going to be an additional kitti semantics dataset released, you can check this: http://panoptic.cs.uni-freiburg.de/

[ruili3]

Thank you for your reply~ I double checked the page and found the kitti model on another branch of the project. Thank you for the other link of the extra annotated dataset, it's really awesome!