I have to question. I can't find any answers online therefore I'm going to ask them here.

  1. Is Image Augmentation in the context of Object detection always meaningful? I have 100 images of a object which looks almost the same in every picture. Same background, same color, same shape. The validation data and the expected data in the future will look similar. So for me there is actually not really the need of generalization. If I add augmented images to the train data, could they have a negativ influence because the model tries to learn features (e.g. different colors, shapes..) it actually does not need to learn, because they don't occur in the real-life data?

  2. I know the concept of online learning or stochastic gradient descent. In comparison to batch learning, we adjust the weights after one training element. Imagine I have a trained model and with time a get some new training elements, but only one per time. So I want to fine tune the existing model with that new element. What I don't understand, how many epochs/iterations should there be with that single element? Just one weight-adaption? 10, 100?

Thanks for any help! Cheers!


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.