Move and Act: Enhanced Object Manipulation and Background Integrity for Image Editing

Setup

You can build the environment as follow:

conda env create -f environment.yaml
conda activate ldm

Usage

In order to edit the action of an object in an image and specify the generation position of the edited object, you can use run_mna.py. For example:

python run_mna.py --img_path './data/0040.jpg' --cond_path './condition/0040.json'

Or you can specify that the edited object is generated at its original location:

python run_mna.py --img_path './data/0012.jpg' --cond_path './condition/0012.json'

The direcory structure of editing results are as follows:

outputs/
|-- text prompt/
|   |-- 42.png 
|   |-- 42_bbox.png 
|   |-- 1.png
|   |-- 1_bbox.png 
|   |-- ...