Testing BIGAN (Adversarial Feature Learning) for State Representation Learning
This is a PyTorch implementation of a BIGAN Network described in the paper "Adversarial Feature Learning" by J. Donahue, P. Krahenbuhl, T. Darrell.
This program will be tested on datasets from "Learning State Representations with Robotic Priors" (Jonschkowski & Brock, 2015), https://github.com/tu-rbo/learning-state-representations-with-robotic-priors
Usage:
python3 main.py [-h] [--dataset {mnist,robot_world}]
[--dataset_path DATASET_PATH] [--gpu_mode GPU_MODE]
[--save_dir SAVE_DIR] [--result_dir RESULT_DIR]
[--log_dir LOG_DIR] [--epoch EPOCH] [--batch_size BATCH_SIZE]
[--lr LR] [--beta1 BETA1] [--beta2 BETA2] [--slope SLOPE]
[--decay DECAY] [--dropout DROPOUT] [--network_type {FC,CNN}]
[--z_dim Z_DIM] [--h_dim H_DIM]
Pytorch implementation of BIGAN
optional arguments:
-h, --help show this help message and exit
--dataset {mnist,robot_world}
The name of dataset
--dataset_path DATASET_PATH
--gpu_mode GPU_MODE
--save_dir SAVE_DIR Directory name to save the model
--result_dir RESULT_DIR
Directory name to save the generated images
--log_dir LOG_DIR Directory name to save training logs
--epoch EPOCH The number of epochs to run
--batch_size BATCH_SIZE
The size of batch
--lr LR Learning rate
--beta1 BETA1 for adam
--beta2 BETA2 for adam
--slope SLOPE for leaky ReLU
--decay DECAY for weight decay
--dropout DROPOUT
--network_type {FC,CNN}
Type of network (Fully connectec or CNN)
--z_dim Z_DIM The dimension of latent space Z
--h_dim H_DIM The dimension of the hidden layers in case of a FC
network
Example:
python3 main.py --network_type FC --dataset robot_world --epoch 50 --z_dim 2