search

SimonVandenhende / Multi-Task-Learning-PyTorch

PyTorch implementation of multi-task learning architectures, incl. MTI-Net (ECCV2020).
Other
752 stars 114 forks source link

ResNet-50 single task baseline hyperparameters #5

Closed zergzzlun closed 3 years ago

zergzzlun commented 3 years ago

Hi there, thank you for open-sourcing your amazing work.

I trained ResNet-50 single-task baseline for segmentation task using config file "configs/nyud/resnet50/semseg.yml". However, I can get at best 40.4 mIoU on NYUD-v2 dataset, which is lower than 43.9 reported in the survey. Can you please provide the hyperparameter used in the survey or some hint to reach 43.9 mIoU?

Thanks in advance.

SimonVandenhende commented 3 years ago

Hi. Thanks for the interest. Please refer to issue #1 . The results on NYUD were obtained using different augmentations, compared to the ones I opensourced here. If the issue still persists, please contact me.
As promised in issue #1. I will work to integrate this into the code base this month, and try to opensource the pre-trained models.

SimonVandenhende commented 3 years ago

I am working to fix the issue this week.

SimonVandenhende commented 3 years ago

The issue should be fixed. Please refer to #1 for more information.

zergzzlun commented 3 years ago

The issue should be fixed. Please refer to #1 for more information.

Many thanks.

zergzzlun commented 3 years ago

The issue should be fixed. Please refer to #1 for more information.

It seems the correct version of NYUDv2 dataset cannot be downloaded right now. The URL is '/TO/BE/DEFINED/' in the new version of dataloader: link.

SimonVandenhende commented 3 years ago

Should be fixed now. I forgot to commit the latest change to download the data automatically. Sorry for the inconvenience.

zergzzlun commented 3 years ago

Should be fixed now. I forgot to commit the latest change to download the data automatically. Sorry for the inconvenience.

Thanks for the update.