I Didn't Want to Upload My Model Just to Convert It
I made a tool to convert my trained models into deployment formats, fully on my machine, so I never have to upload a private model just to export it.
Here's the thing that annoyed me. You train a model, it works, and then you have to convert it to run somewhere else. A phone, a Jetson, a browser. That means turning your YOLO .pt or RF-DETR .pth into ONNX, TensorRT, CoreML, or TFLite.
Sure, you can do this with scripts. But I got tired of writing the same script again and again. And every time I moved to a new setup, the script broke. Not fun.
The bigger problem for me was privacy. A lot of tools want you to upload your model to convert it. I didn't want to do that. My model is private. I shouldn't have to put it on the internet just to change its format.
So I made Vision Export Studio. You load your model, pick YOLO .pt or RF-DETR .pth, choose the format you want, and it converts it. Everything runs on your machine. Nothing gets uploaded.
If you've had the same problem, it's on GitHub.