I cannot understand why distorting an image, e.g flipping it, increasing the gamma intensity would somehow increase the accuracy on neural network.

Within my situation, I am Using a CNN to detect if dogs are present in an image, and I was recommended to add distortion.


If we have the input of an image the CNN detects certain features and patterns that together make up a dog. If we only feed it images of a dog looking to the right it will not know that dogs can also look to the left, so it will not recognize the dog in a flipped image.

This increases the generalization potential of your network. The distortion technique does something similar but in a different way. An image is taken and slightly changed. For humans it is still clearly recognizable as a certain object, but the network gets a different input. This way it gets to learn slight variations of the same problems to see these are essentially the same. This once again increases the generalization potential.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.