Generating images from scene graphs using conditional control of diffusion models.
- Right now we use a modified version of sg2im for scene layout generation
- Scene graphs are stored in data.json
- Scene layout generation (bounding boxes and semantic maps) is done in prepare.ipynb
- ControlNet.ipynb creates images from semantic maps
- GLIGEN.ipynb creates images from bounding boxes
- evaluate.ipynb measures insception score and FID score