search

mit-han-lab / spvnas

[ECCV 2020] Searching Efficient 3D Architectures with Sparse Point-Voxel Convolution
http://spvnas.mit.edu/
MIT License
587 stars 109 forks source link

MinkUnet reproduction w/ much larger epoch #112

Closed kimbss470 closed 1 year ago

kimbss470 commented 1 year ago

Hi, Thank you for sharing your work!

When I've tried to train MinkUet@114GMACs model with 100 epochs (other hyperparameters & training configurations are default ones) , I got mIoU 65.28% on validation set and 62.95% on test set.

I just wanna know why you didn't use larger epochs even if it incurs better results.

zhijian-liu commented 1 year ago

A longer training schedule can sometimes lead to improved performance. However, it also takes much longer training time. In our original paper, we intend to keep the same number of training epochs for fair comparisons.

deeplearning666 commented 1 year ago

Hello, When I download the pre-trained model, there is an error, like "Page Not Found. The page you are looking for doesn't exist or has been moved". I can not download the provided model, could you please share the provided model with me? Thanking you! @kimbss470 @zhijian-liu