Clarification on ImageNet pre-trained weights under Open Source / Commercial rules

DaT Parkinson's Prediction Challenge
1

Hi DrivenData Team,

Could you please clarify the stance on using ImageNet pre-trained weights as external data for this challenge?

According to the competition rules:

  1. Winning solutions must be available under an open-source license (MIT) that does not prohibit free commercial use.

  2. External data (including pre-trained models) must be used pursuant to a valid license permitting use without requiring third-party commercial approvals.

Standard deep learning backbones (such as PyTorch torchvision or timm models) are frequently initialized with ImageNet pre-trained weights. While the framework code is open source (Apache 2.0 / BSD), the original ImageNet dataset specifies non-commercial research use.

To ensure compliance ahead of time:

  • Are standard pre-trained ImageNet weights permitted as external data for model initialization in this challenge?

  • Do solutions using ImageNet-initialized backbones satisfy the commercial-use requirements for prize eligibility?

Thank you for your guidance!

2

Hi - good question!

Generally, external models are eligible for prizes as long as the model or weights themselves are permissively licensed (such as MIT or Apache 2.0). The license needs to allow derivative works to be used and distributed commercially, so it’s worth keeping an eye on the licenses attached to the specific weights you use. If your solution places, we’ll verify against those terms then.