Using Generative Adversarial Networks (GANs) to produce awesome looking fantasy maps.
Implementation details: A simpler version (i.e., removal of the conditional image generation functionality) of the AE-GAN I implemented in my Pokemon Sprite Generator project. For more details, definitely check out that repo! I'm working on having this GAN though scalable to larger sizes (originally 96x96 - want to get up to 256x256.)
Below is a rough architecture flow diagram of what's happening during training. In this case x and r are images, z are latents. We can compute reconstruction losses between recon_loss(x, r') and generator GAN losses from disc_image(r), disc_image(r'), disc_latent(z'), and disc_latent(z''). Conversely, we can compute discriminator losses from disc_image(x), disc_image(r), disc_image(r') and disc_latent(z), disc_image(z'),disc_image(z'').
[NOTE] Results coming soon! Currently training a 128x128 version of the net, and things are looking good.
-
I used the Selenium IDE to build the dataset of fantasy maps. Credits to @mewo2, who built the map generator I used to construct the dataset. The dataset is available on Google Drive link. The dataset contains 4000 images.
-
Install necessary python packages. If doing training, I'd highly reccomend using a GPU. These are the package versions I was using, but I'm sure it would work for other combinations as well.
python==3.6.12
torch==1.8.1+cu111
torchvision==0.9.1+cu111
PIL==8.0.1
tensorboard==2.5.0
numpy==1.19.4
- (Optional) Download the pretrained model:
cd fantasy-map-generator/
mkdir pretrained
cd pretrained/
# TBD
wget <URL>
The program is run via the command line. There are two modes, train or sample, which we'll outline in more detail below. For now, here is the full list of command line options:
python main.py -h
usage: main.py [-h] [--mode {train,sample}] [--save_dir SAVE_DIR]
[--load_dir LOAD_DIR] [--use_gpu] [--root_dir ROOT_DIR]
[--csv_file CSV_FILE] [--batch_size BATCH_SIZE]
[--learning_rate LEARNING_RATE] [--num_epochs NUM_EPOCHS]
[--types TYPES]
pokemon-sprite-generator options
optional arguments:
-h, --help show this help message and exit
--mode {train,sample}
run the network in train or sample mode
--save_dir SAVE_DIR path to save model, logs, generated images
--load_dir LOAD_DIR path to model to load
--use_gpu if set, train on gpu instead of cpu
--root_dir ROOT_DIR path to the training data
--csv_file CSV_FILE path to the training data
--batch_size BATCH_SIZE
batch size
--learning_rate LEARNING_RATE
learning rate
--num_epochs NUM_EPOCHS
number of epochs
--types TYPES pokemon types, comma seperated
To train the network from scratch, I'd highly recommend using a CUDA-enabled GPU. It took me about ____ TBD. I'd also recommend keeping the default network hyperparameters. So, your command to train might look like:
python main.py --mode train --save_dir logs
If you'd like to generate a random map using the pre-trained model, make sure to first download the pre-trained model weights from step 3 of the Setup section. To generate a map, your command might look like:
python main.py --mode sample --save_dir logs --load_dir pretrained
The generated map will be saved to the save_dir specified.
TBD - training curves, visualizations, etc.
Some inspirations for this work!