I found that CIFAR dataset is 32px*32px, MIT 128px*128px and Stanford 96px*96px.
if I would like to use different resolutions, can I just resize them to the smaller?
https://www.quora.com/To-what-resolution-should-I-resize-my-images-to-use-as-training-dataset-for-deep-learning
It depends of the DL library you are using. Some of them have a input resolution of 256x256 pixels. Furthermore, it also depends of the method you are using, overall convolutional methods for image classification require big resolutions.
If you resize images to small resolutions you may lost important information depending of the content of the images.
Learning should be performed with resolutions similar to later processing. Otherwise the neural net cannot learn what you intend to.
It is not relevant how many pixels the images are containing but the grade of detail in the images.
I also think it's better to look at the average size of all the images, taking an average size and image rotations can help you make a good training.
- Badges
- Science topic