I'm creating a 'convolutional neural network' to detect NSFW (Not safe for work) or 18+ images. But I do not know the ratio and size of my dataset. I hope to put together about 70,000 images with 50% sfw and 50% nsfw. In which I will train the network with 50,000 images and validate it with 20,000.
What is the best proportion for me to create this dataset?