I would like to add some more samples to my dataset which consists of two parts: 1. image and 2. numerical data. For each image in the dataset there is its corresponding numerical data as well. If it were to image only classification, I would have used cropping augmentation method (others wouldn't not suit the dataset). However, it is now a multiple input to one output (binary classification) task, I'm struggling to come up with a better solution.

Could someone name some of the techniques I can apply to such a type of classification?

One example I can think of is this. Since I can't apply other augmentation methods except cropping, I may increase one sample of data by cropping the center, left and right parts (by preserving ~90% of original image).

(cat image and its numerical data were made up for explanation) enter image description here

If this type of image augmentation is still sensible, I would just broadcast the same number of numerical samples (namely, 3 times) as one image's numerical data.

enter image description here

What do you think of this way of data augmentation?


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.