GLIGEN: Open-Set Grounded Text-to-Image Generation (CVPR 2023)

Fork of the original GLIGEN repo for development purpose. Please read the original repo to prepare the environments.

Here's what I've done.

Download and process all data according to their instructions at /shared/patrickwu/GLIGEN. This includes executing the tsv_split_merge.py program (which I've fixed a bug) and updated the path in dataset/catalog.py.
Download the pretrained sdv1.4 checkpoint at /shared/patrickwu/GLIGEN/DATA/sd-v1-4.ckpt (from) as GLIGEN needs it when running their code.
Run the code: python main.py --name=trial_batch --yaml_file=configs/alldata_boxtext2img.yaml

Name		Name	Last commit message	Last commit date
Latest commit History 60 Commits
DATA		DATA
configs		configs
dataset		dataset
demo		demo
docs		docs
env_docker		env_docker
figures		figures
grounding_input		grounding_input
inference_images		inference_images
ldm		ldm
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
SD_input_conv_weight_bias.pth		SD_input_conv_weight_bias.pth
color150.mat		color150.mat
convert_ckpt.py		convert_ckpt.py
distributed.py		distributed.py
gligen_inference.py		gligen_inference.py
inpaint_mask_func.py		inpaint_mask_func.py
main.py		main.py
projection_matrix		projection_matrix
trainer.py		trainer.py
tsv_split_merge.py		tsv_split_merge.py

Provide feedback