Open isunLt opened 2 years ago
I also trained gasn on semantickitti and nuscenes with four GPUs but only get 67.7 on semantickitti and 70 on nuscenes. I keep the same hyper-parameters on the two datasets. @isunLt , would you please share the hyper-parameters on semantickitti. Thanks!
I keep the default hyper-parameters as the author. The training performance may vary a little on different environment.
oh, thanks!
Checklist
First of all, thanks for your great work! I wonder will you release the training code for the nuScenes dataset? I tried to train gasn on nuScenes but only get 69.0 at mIoU (it is strange as I get 68.3 on SemanticKITTI). Is there any different between the training configs of the two datasets? For example, batch size, epoch, grid size, the channel of hiden layer.