I am working on image processing project. I see that it's recommended to resize the image in the data pre processing step. How do I decide what size to to resize it to? The tutorial I am following resizes it to 128.
It actually depends on your deep learning model and your GPU.
So the rule of thumb is use images about 256x256 for ImageNet-scale networks and about 96x96 for something smaller and easier. I heard that there is also people who train using 512x512 on kaggle site too.
If you train fully convolutional networks like Faster RCNN you can take much bigger images (say 800x600) because you have batch size = 1, but overall once again it really depend on what deep learning model you are using and the resources you have at hand.