Prediction on https://app.roboflow.com/ under "Try this Model" differs from when using Serverless Hosted API

Uploading an image on Try this Model on app.roboflow.com provides different prediction values. What is being done behind the main app when we upload an image to try the model?

Using detect.roboflow.com, the prediction output also differs too much from the main app.

Some parts of the main app use tensorflow js to run the image inference locally in the browser. This can differ in results from other inference methods!

Thanks for the explanation @Lake. May I check if the image inference code using tensorflow is opensource or viewable somewhere?

Our inferencejs code is not open source, but you can see an example of how it is used here! GitHub - roboflow/inferencejs-react-example