cv-ip / inr-gan Goto Github PK

Official implementation of Adversarial Generation of Continuous Images

Disclaimer

Right now, the repo is being rewritten/refactored from firelab to pytorch-lightning, so you may like to wait a week or two for the update before starting building it.

INR-based GAN

Generating images in their implicit form.

To run the reconstruction model, you should first install firelab library:

pip install firelab

To run the experiment on a single GPU, use the following command:

CUDA_VISIBLE_DEVICES=0 python src/run.py -c configs/inr-gan.yml --config.dataset lsun_256

To run a multi-gpu training, we use horovod which is launched via:

horovodrun -np NUM_GPUS --mpi-args=--oversubscribe python src/run.py -c configs/inr-gan.yml --config.distributed_training.enabled true --config.dataset lsun_256 --config.hp.batch_size BATCH_SIZE