Why preprocess images before image training in deep learning

Data Enhancement That is, increase the amount of data, the original figure of this category is too small, how to do? The existing map flip, pan, fogging, de-drying and other methods to increase the sample size, easy to train. For example, your current images are taken during the day under the sunny bike, you use to train the model, but let you detect when there is a picture of foggy road conditions during the day. It might not recognize the bike coming. But if you have the picture in your training set after you fogged it, you trained for that situation and might be able to recognize it 。。。。。 It learns what you feed it. The more comprehensive the situation you give it the better it learns.