LostGANs: Image Synthesis From Reconfigurable Layout and Style

This is implementation of our paper Image Synthesis From Reconfigurable Layout and Style and Learning Layout and Style Reconfigurable GANs for Controllable Image Synthesis

Network Structure

Installation

Check INSTALL.md for installation instructions.

1. Download pretrained model

Download pretrained models to pretrained_model/

2. Train models

python train.py --dataset coco --out_path outputs/

3. Run pretrained model

python test.py --dataset coco --model_path pretrained_model/G_coco.pth --sample_path samples/coco/

Results

Compare different models

Multiple samples generated from same layout

Synthesized images and learned masks for given layout

Contact

Please feel free to report issues and any related problems to Wei Sun (wsun12 at ncsu dot edu) and Tianfu Wu (tianfu_wu at ncsu dot edu).

Reference

• Synchronized-BatchNorm-PyTorch: https://github.com/vacancy/Synchronized-BatchNorm-PyTorch
• Image Generation from Scene Graphs: https://github.com/google/sg2im
• Faster R-CNN and Mask R-CNN in PyTorch 1.0: https://github.com/facebookresearch/maskrcnn-benchmark