If I have a new ‘test set’ with new images I have just gathered, how can I add it to the existing vision learner ? Actually I want to run the model with the new images, I think if I tried just learn.model([Img]) then it will not work because we need to do some preprocessing with transformation and normalization. I would be very appreciated if anyone can suggest me a way to do it ?
Project I am working on:
I need a model that can add new category regularly. I found this link here proposes ‘Content-based image retrieval’ to solve the problem. In short, in order to not retrain whole model, we can compare the features vectors to retrieval which image is closest to the test image. We can mask the feature vector with binary mask (value > 0 or < 0) than using hamming distance (just the sum of the bits difference) to evaluate the similarity. You find here the paper for this technique: Deep Learning of Binary Hash Codes for Fast Image Retrieval
Now each time I have a new category, I need to call the model on it to find the feature vector