Hello @discourse-admin,
Starter notebook mentions SegFormer as a possible model for the task, but it has non-commercial license. Is it OK? Thanks in advance!
Hi @vecxoz
Great question about the SegFormer license. In the starter notebook, we provided SegFormer as a potential example for a way to solve this challenge. However the non-commercial license in the repository precludes use for the challenge. But it might be a great way to research other potential ideas built on that architecture.
Onward Team
1 Like
Thank you for the confirmation, @discourse-admin. Of course studying SegFormer implementation is beneficial in many ways regardless of its license terms.
Hello @discourse-admin,
Please, could you confirm another non-trivial case of licensing. Well known segmentation_models_pytorch framework is distributed under MIT license. But among its encoders there is a Mix Vision Transformer which is a SegFormer backbone, called mit_b0, …, mit_b5. Given that it is a derivative work from SegFormer, I think it is safe to assume that this encoder has non-commercial license and hence it is also forbidden in the competition. Is this correct? Thanks in advance.
Hi @vecxoz
You are correct, since the Mix Vision Transformer is a derivative work of the SegFormer backbone which has a non-commercial license, it should not be used in this challenge.
Good luck with training
Onward Team
1 Like
Thank you very much!