Clarification on GPLv3 License When Using Quantized YOLOv7 Weights Without Source Code

[hanshin-ps]

@WongKinYiu @AlexeyAB @spacewalk01 @raymondben @alexeysi

Subject: Clarification on GPLv3 License Scope When Using YOLOv7 Training Code Only

Hi authors

Hello, and thank you for your great work on YOLOv7.

I would like to ask a question regarding the GPLv3 license and how it applies in a specific case.

In my project, I trained a model using the official YOLOv7 training script (train.py) along with my own dataset. After training, I exported the resulting model to the ONNX format and then quantized it into another format suitable for deployment on an embedded device.

At runtime, this quantized model is used without any part of the YOLOv7 source code. Inference is performed entirely using our own custom inference code, and we do not use any code from the YOLOv7 repository for model loading, preprocessing, or postprocessing.

In this scenario, does the GPLv3 license still require us to open our source code or the model files, even if no YOLOv7 source code is used in deployment?

I would really appreciate your clarification on this matter.

Best regards,