Complete A.I. Art Generation Course - Beginner 2 MASTER

Explore how a generative adversarial network uses a generator and a discriminator to produce high-quality images from noise, guided by a loss function and iterative training.

Begin the course by building a simple gan to generate handwritten digits, then open the Google Colab link to start coding and watch the next video.

Set up a gpu runtime, load amnesty digits or fashion amnesty datasets, and use the five import statements as neural network building blocks with Nampai to generate images.

Builds a generator that converts latent dim input into a 28x28x1 image using a sequential neural network with dense layers, leaky relu, batch normalization, and tanh.

Build the discriminator for the GAN by flattening the input, adding a 512-unit dense layer with leaky relu, and using a sigmoid output to classify real versus fake images.

Connect generator and discriminator into a full gan, configure binary cross entropy loss with the Adam optimizer, freeze the discriminator during training, and prepare for training.

train a gan by loading and normalizing the data, setting epochs and batch size, and alternating discriminator and generator training to produce and save images for GIFs.

Explore building a better gan by importing python packages, loading the Safar 10 dataset, and using an alternate generator architecture with transposed convolutions to produce rgb images.

Tune the neural network variables to match the 50,000 image, 32x32 rgb dataset. Store these parameters to prepare for building the generator in the next lecture.

load folder of Bob Ross paintings as a dataset, compress to zip, upload to Google Colab or Drive, unzip, and load images into a numpy array to train a model.

Resize all dataset images to 64 by 64 by 3, save them in a resized images folder, and ensure correct path handling to feed into the neural network in colab.

Adjust the generator and discriminator to handle 64 by 64 by 3 inputs, explore scaling to 128 by 128, verify shapes, and prepare to load Bob Ross paintings for generation.

Load dataset images from a directory, create an array, normalize data, and train the model with a tuned batch size while saving weights and monitoring discriminator and generator losses.

Upload and unzip painting datasets from 1k to 15k images, adjust paths, and reshape to 128 by 128 for neural network input; then train the generator and discriminator.

Explore long short term memory (LSTM) neural networks that excel at processing time sequences to power machine translation, text generation, music generation, and predicting bitcoin prices.

Learn to import tensorflow and two natural language processing functions to turn Shakespeare text into numbers, download the dataset, and tokenize this text for a neural network to generate text.

Tokenize the text to convert each word into a corresponding number and build a word index. Lowercase the corpus and split it on new lines to prepare data for steps.

Tokenize the corpus and map words to numbers, creating input sequences with the graham sequence targets and pad them to a fixed length for neural network training.

Generate a Star Wars script with a bidirectional neural network, using word-level tokenization, padding, and return sequences, trained on a Star Wars text file to improve accuracy.

Learn how to tokenize and encode a blog title, then generate a 200-character blog with GPT-2, exploring beams, no-repeat, and early stopping.