about validate and inference

[Huntersxsx]

Hello, I notice that you use validate function during training while using inference function during testing, and the calculated results of these two functions for the same checkpoint are different.
The main difference between these two functions' implementation is that 'validate function' calculates IoU among a batch while 'inference function' calculates IoU among different sentences regarding the same image. However, I find that you use mean IoU as the evaluation metric, which is the average IoU across all test samples. I am confused that why this difference will lead to different mIoUs?
Is there something I misunderstand or neglect?
Why use two different functions during validation and inference?
Looking for your reply~