00:00

Understanding Generative Adversarial Networks (GANs)

Learn about Generative Adversarial Networks (GANs), a type of deep learning model that consists of a generator and a discriminator. Explore key components such as convolutional layers, Leaky ReLU activation, batch normalization, and transposed convolutions. Discover how GANs can be used for tasks like image generation and how to implement GANs using TensorFlow. Dive into training techniques, loss functions, and optimization methods for GANs. Experiment with GANs on the MNIST dataset and explore conditional generation and image interpolation.

uliarte
Download Presentation

Understanding Generative Adversarial Networks (GANs)

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Project 5: Generative Adversarial Network Zhifei Zhang

  2. Generative Adversarial Network (GAN) https://github.com/hindupuravinash/the-gan-zoo

  3. Generative Adversarial Network (GAN)

  4. Generative Adversarial Network (GAN) Binary Classifier: Conv, Leaky ReLU, FC, Sigmoid New components: Transposed convolution, Batch Normalization https://github.com/PramodShenoy/GANerations

  5. Transposed Convolution (Deconvolution) Stride=1 Pad=Valid Stride=2 Pad=Valid Stride=2 Pad=Same Output Conv Input Output Deconv Input https://github.com/vdumoulin/conv_arithmetic

  6. ReLU vs. Leaky ReLU (LReLU) Used in Generator Used in Discriminator

  7. Batch Normalization (BN) Saturation area Zero gradient https://towardsdatascience.com/batch-normalization-in-neural-networks-1ac91516821c

  8. Generative Adversarial Network (GAN) Reshape, FC, sigmoid FC, BN, ReLU Reshape Conv (kernel 5x5, stride 2) BN, LReLU (slope 0.2) Deconv, tanh/sigmoid Deconv (kernel 5x5, stride 2) BN, ReLU https://github.com/PramodShenoy/GANerations

  9. GAN on MNIST Deconv Tanh/Sigmoid Conv, BN, LReLU Reshape, FC, BN, LReLU Conv, BN, LReLU FC, Sigmoid FC, BN, Reshape Deconv BN, ReLU 100 7x7x16 14x14x8 28x28x1 7x7x16 256 1 14x14x8 Generator Discriminator

  10. Conditional GAN on MNIST One-hot label Deconv Tanh/Sigmoid Conv, BN, ReLU Reshape, FC, BN, ReLU Conv, BN, ReLU FC, Sigmoid FC, BN, Reshape Deconv BN, ReLU 100 7x7x16 14x14x8 28x28x1 7x7x16 256 1 14x14x8 Generator Discriminator

  11. TensorFlow Implementation of GAN Run the demo code: (Linux or MacOS) $ git clone https://github.com/carpedm20/DCGAN-tensorflow $ cd DCGAN-tensorflow $ pip install tqdm (if you do not have this package) $ python download.py mnist $ python main.py --dataset mnist --input_height=28 --output_height=28 --train

  12. TensorFlow Implementation of GAN • Input: z, image, (label) Training: • Network: D, G for epoch • Loss: D, G for batch • Optimizer: D, G Update D Update G

  13. https://devblogs.nvidia.com/photo-editing-generative-adversarial-networks-1/https://devblogs.nvidia.com/photo-editing-generative-adversarial-networks-1/

  14. TensorFlow Implementation of GAN Testing: • Random Generation • Conditional Generation • Interpolation

More Related