How to evaluate the performance of a trained model on another dataset?

[Gratia2533]

Hello everyone,
I'd like to ask, after training, if I want to use this model to extract images from another dataset, how can I know the performance of the extraction? (e.g., accuracy) I have the groundtruth of this dataset.

There's another small question I have. In the Train section on Topaz GUI, I saw a hint regarding the CNN model options as follows:

"Model type to fit (options: resnet8, conv31, conv63, conv127) (type: string)
Note: Your particle must have a diameter (longest dimension) after downsampling of:
70 pixels or less for resnet8
30 pixels or less for conv31
62 pixels or less for conv63
126 pixels or less for conv127"

However, in other discussions, I've seen the author answer questions about the radius: training radius < actual radius ≈ extraction radius. So, when choosing which model to use, which radius should I use for better performance?

Thank you very much.