In the rules we have the following point: “hereby grants to Sponsor and Sponsor Admins a non-exclusive, irrevocable, royalty free, world-wide right and license to the Entry for any purpose”. But the baseline uses SegFormer which is non-commercial. Are participants allowed to use this model in their solutions? Thanks in advance.
2 Likes
Hi @dipam, please take a look at this thread.
Also I believe another one is important as well. Thanks in advance.
Hi! That disclaimer applies to code and not methods themselves. If you are planning to use external code, please check that the license is compatible with this. Regarding the baseline using SegFormer, we employ a different implementation than the one from NVLabs, whose license you link, in particular we use the MMSegmentation library under a Apache 2.0 license, this license is compatible with the challenge rules and therefore you could use this specific implementation of SegFormer for your submission.
Hi @edl, many thanks for your answers. But if I understand it correctly, the license should persist across all derivative works. MMSegmentation has the following mentions:
https://github.com/open-mmlab/mmsegmentation/blob/master/LICENSES.md
https://github.com/open-mmlab/mmsegmentation/blob/master/mmseg/models/decode_heads/segformer_head.py
I’m not trying to be picky. It is a computationally intensive model so I just want to make sure it is allowed before I start to train it. Thanks!
Hi! Good catch!
It is alright to use external code under the NVidia license for submissions, as it would allow us to reproduce your training in our system. However all the custom code you generate should comply with the rules.
Also if you attach external code to your submission, please include all corresponding licenses. We recommend you link and provide instructions on how to use the external code needed to reproduce your submission.
Looking forward to your participation!
1 Like