I have read some online tutorials about using Siamese network for identification purpose but I’m confused about one thing:
In the training and validating process, a Siamese network often calculates loss and accuracy values based on batch size. So, even if the generalised model can achieve a good result, the application in reality will be different because we have to match an image against hundreds or thousands of images, not just some images within the scope of batch size. In this case, the accuracy is believed much lower.
Therefore, if a research or online tutorial just reports the accuracy at the training level (i.e. 95% or something), I’m concerned that how that result reflect the practical picture in term of application.
Looking forward any idea or advice on this concern.