MinkUNet18 and MinkUNet34

[YilmazKadir]

Hi,
It looks like MinkUNet18 and MinkUNet34 were confused with each other.

NCLR/downstream/networks/backbone/minkowski_engine/minkunet.py

Line 838 in 88bb822

class MinkUNet34(MinkEngMinkUNet):

You can check from the original repository: https://github.com/NVIDIA/MinkowskiEngine/blob/02fc608bea4c0549b0a7b00ca1bf15dee4a0b228/examples/minkunet.py#L193

Also, which model did you use for training?

[Eaphan]

We have several configurations for experiments on downstream tasks. Depending on the configuration, we utilize different networks for training, specifically based on the value of network.backbone.

[YilmazKadir]

Thanks for the clarification, but I think there’s a misunderstanding. What I meant is that the naming of MinkUNet34 and MinkUNet18 in your codebase seems to be swapped — the architectures themselves appear mislabeled and should be corrected.

NCLR/bevlab/backbones/minkunet_me.py

Line 855 in 88bb822

class MinkUNet34(MinkEngMinkUNet):

and

NCLR/downstream/networks/backbone/minkowski_engine/minkunet.py

Line 838 in 88bb822

class MinkUNet34(MinkEngMinkUNet):