Create images with randomized size?

[Nestak2]

Hi @mdbloice, thanks for the Augmentor code! I am creating a pipeline, that has to create eventually images with randomized size, for example: I have a single ground truth image of size 100pixel x 100pixel and after running the pipeline I would like to have images scaled with a RANDOM SCALE in the interval [1,2], so that output images have sizes 173x173, 189x189, 114x114, 133x133 etc. The closest to this functionallity I have found was crop_random, when the argument randomise_percentage_area is set to "True". But in this case percentage_area is an UPPER bound, while I need it to be a LOWER bound. Of course, for creating N output images I could run N times the pipeline, every time changing the scale factor, but I hope there is a handier alternative. Therefore my question: 1.) is there a function in Augmentor that allows for output images with randomized size/scaling? 2.) alternatively, can percentage_area be set to be the LOWER limit, instead of the upper, when randomise_percentage_area=True?

Thanks for an answer!

[mdbloice]

Hi @Nestak2, thanks for making the issue and sorry for the delay in replying, I have not had much time for Augmentor recently.

Yeah, that would be a good feature to have. Right now I cannot think of a way you might be able to do what you are asking with the current operations as they exist now...

However, I can add this functionality in the next version, I think it would be useful. Until then though, I can't really help out any more.

If I can think of a way to do what you are asking as is, I'll post it here.

[Nestak2]

Thanks for replying! I actually managed to make my own pipeline out of opencv functions. Though it might be useful for others in the future to have the functionality in Augmentor.

[mdbloice]

Yeah OK good that you sorted it, thanks for opening the issue I will add it as a feature when I get the chance.

[cahya-wirawan]

Hi, I would like also to have functionality to set lower and upper bound like in torchvision as torchvision.transforms.RandomResizedCrop(size, scale=(0.08, 1.0), ratio=(0.75, 1.3333333333333333), interpolation=2). Thanks for the great library anyway.