Скачать с ютуб I created an AI that generates anime faces! в хорошем качестве

I created an AI that generates anime faces!

I've noticed few generated samples by generator which are very blurry, and that is because in our true data we do have few blurry anime faces, so I think that just shows how well the generator was able to understand this dataset :) Okay, so points to keep in mind while implementing:- 1. Replace any pooling layers with strided convolutions (discriminator) and fractional-strided convolutions (generator). 2. Use batchnorm in both the generator and the discriminator. 3. Remove fully connected hidden layers for deeper architectures. 4. Use ReLU activation in generator for all layers except for the output, which uses Tanh. 5. Use LeakyReLU activation in the discriminator for all layers. 6. Initialize weights of conv and deconv layers and gamma variable of batchnorm layer with normal distribution with mean 0.0 and std 0.02, and beta variable of batchnorm layer with 0.0 7. No batchnorm in first and last block of discriminator and generator respectively. 8. Setting up beta1 value for adam optimizer = 0.5 or less. (Not too less though) 9. Normalize true data from range -1.0 to 1.0 10. Have good patience. One correction, in paper they are using uniform distribution for providing random noise to the generator, which I missed. GitHub Repository: https://github.com/braindotai/Fake-An... A start ⭐ would be saluted. 0:00 Intro 0:10 Final results we'll get 0:32 Paper introduction 2:08 Understanding paper 7:17 Starting implementation. 7:52 Setting up Kaggle 8:17 Downloading dataset to Google Colab 8:34 Setting up weights and biases 8:52 Importing required libraries 9:33 Creating dataset iterator 10:03 Visualizing a sample from dataset 10:23 Creating dataloader, and visualize few samples from a batch 10:57 Creating the discriminator 12:01 Summarizing the discriminator 13:03 Creating the generator 13:51 Initializing parameters 14:20 Defining losses for generator and discriminator 14:39 Defining optimizers for generator and discriminator 15:15 Defining helper functions for saving and loading checkpoints 15:44 Defining noise generating function 15:53 Performing initial save and load 16:08 Setting up the weights and biases for this run. 16:39 Coding the training part 17:38 Starting training and waiting... 18:06 Visualizing final results 19:40 Visualizing image interpolations 20:55 Enjoying the music For more cool projects you can checkout https://www.nevronas.in/