Resize to 224×224 directly or resize to 256 ×256 then crop 224×224?

My images in training set are leaves like this. enter image description here

Its size is 572*108 and my resnet network need 224×224 images as input. I found most of the codes process the images with the second way(resize to 256 ×256 then crop 224×224) so I did that. As a result,parts of my leaves were cut which may influence the effect of leaves classification.Like this: enter image description here

The first way (resize to 256×256 directly) may keep more complete structure.Like this: enter image description here

I am worrying that if choose the first way I will lose many training images since random crop 224×224 may generate more samples for training.

Which should I choose?

1 answer

  • answered 2022-04-24 16:08 chongkai Lu

    Running two experiments and compare their evalution results is the simplest solution.

    A complete view is not neccessary for model to classify images, so as it to human. On the contrary, learning from cropped image normally can improve the generalization capacity of a model.

How many English words
do you know?
Test your English vocabulary size, and measure
how many words do you know
Online Test
Powered by Examplum