perturbed-seg
Code for NeurIPS 2019 paper Emergence of Object Segmentation in Perturbed Generative Models
Stars
30
Ecosystems:
Python
Emergence of Object Segmentation in Perturbed Generative Models
Code for NeurIPS 2019 paper Emergence of Object Segmentation in Perturbed Generative Models.
Usage
Tested with PyTorch version 1.2.0.
Download LSUN object dataset.
Sample train command for the layered generative model:
python train.py --max_size 128 --location_jitter 0.125 --min_mask_coverage 0.25 --mask_alpha 2.0 --mixing -d lsun /path/to/lmdb/ --bg_contrast_jitter 0.3 --org_to_crop 1.125 --sched
Credits
StyleGAN implementation used in this repository is based on this PyTorch port by rosinality.