We put values of all pixels per channel basis in single row of csv. Details Uses 12 passes at test time. Improving neural networks by preventing co-adaptation of feature detectors. But the major point to note was that the majority class in the dataset was around So, I moved forward with this approach of using a pre-trained Top nn art models and the next step was to fine tune my VGG16 model to suit Top nn art models problem. On the other hand, access to GPUs is not that cheap.