Project is the transfer learning of VGG-19 on 525 Bird Species.
Table of Contents
This repository is the code for a Generative Adversarial Network (GAN) on the CIFAR-100 dataset. The purpose of this code is to test different hyperparameters and neural network architecture, the results of each run are used to create GIF image files that demonstrate the progress of the network over time with different hyperparameters and architectures as indicated in the filenames.
Required Libraries/Frameworks:
Tensorflow (configured for CUDA), Numpy, and Pandas
Required Hardware:
NVidia GPU
- Clone the Repo
git clone https://github.com/zjshermanburke/GAN_CIFAR_100
Modify hyperparameters and the save file paths as desired and run all cells
Hyperparameters Cell
# Latent Space Dimensions
latent_dim = 100
# Training Hyperparameters
batch_size = 32
epochs = 500000
sample_period = 500 # every 'sample_period' steps generate and save some data
# Optimizer Hyperparameters for Discriminator and Combined Model
# Current Optimizer == AdamW
learning_rate = 0.0002
beta_1 = 0.5
Save file paths
# Image Directories
img_dir = './gan_images'
anim_file = './gan_progression_gifs/AdamW_B5_500k.gif'
sub_img_dir = 'AdamW_B1_0.05_Epoch_500k'
# Model Directories
model_dir = './Models/Gan_AdamW_B50_500k'
d_model_dir = 'disc_AdamW_B50_Epoch_500k'
g_model_dir = 'gen_AdamW_B50_Epoch_500k'
comb_model_dir = 'comb_AdamW_B50_Epoch_500k'
Distributed under the Unlicense license. See LICENSE.txt
for more information.
Zachary Sherman-Burke - LinkedIn - [email protected]
Project Link: https://github.com/zjshermanburke/GAN_CIFAR_100