How to verify the generalization ability of the model

Search before asking

• I have searched the Ultralytics YOLO issues and discussions and found no similar questions.

Question

I use YOLO for object detection. The training data was collected in various backgrounds, and the total number of training targets is around 4,000. I conducted an effectiveness test on a completely new set of data, which included different postures of targets captured under various lighting conditions. The results were generally very good. Does this mean that the model has good generalization ability? In other completely different background, but with similar lighting conditions and similar target postures, can it still correctly detect? To add, the reason why I asked the above questions is that I couldn't test all the application scenarios of the model completely. So I wanted to know if it was possible to verify the generalization ability of the model under limited conditions.

Additional

No response