ObjectRemoval

Based on code from:
https://github.com/StanfordVL/GibsonEnv
https://github.com/StanfordVL/3DSceneGraph
https://github.com/seoungwugoh/opn-demo
https://github.com/JiahuiYu/generative_inpainting
The licenses are included in the above links.

Installation

Follow the instructions in the 4 internal README files to download and install the relevant data, code, models and dependencies.
Save the 3DSceneGraph folder ("verified_graph") under GibsonEnv/gibson/assets.

Usage

Build the panoramic masks for all models with:

cd 3DSceneGraph 
bash load.sh

or for one model with:

cd 3DSceneGraph 
python load.py --model model_name

To run object removal for one model:
Run:

cd opn 
python gibson_inpaint.py --model model_name

The rendered input is in "Image_inputs" and the the inpainted output in "Image_results".

To run object removal for all models:

python inpaint.sh

To reproduce the experiments on Places2:

Download Places2 and save under "generative_inpainting/examples/val_256"

Test on OPN:

cd opn 
python test_places2.py

Test on SN-PatchGAN:

cd generative_inpainting 
python test_places2.py --checkpoint_dir model_logs/release_places2_256

Evaluate:

python eval.py