Awesome

ZONE: Zero-Shot Instruction-Guided Local Editing

teaser

Abstract

Recent advances in vision-language models like Stable Diffusion have shown remarkable power in creative image synthesis and editing. However, most existing text-to-image editing methods encounter two obstacles: First, the text prompt needs to be carefully crafted to achieve good results, which is not intuitive or user-friendly. Second, they are insensitive to local edits and can irreversibly affect non-edited regions, leaving obvious editing traces. To tackle these problems, we propose a Zero-shot instructiON-guided local image Editing approach, termed ZONE. We first convert the editing intent from the user-provided instruction (e.g., ``make his tie blue") into specific image editing regions through InstructPix2Pix. We then propose a Region-IoU scheme for precise image layer extraction from an off-the-shelf segment model. We further develop an edge smoother based on FFT for seamless blending between the layer and the image. Our method allows for arbitrary manipulation of a specific region with a single instruction while preserving the rest. Extensive experiments demonstrate that our ZONE achieves remarkable local editing results and user-friendliness, outperforming state-of-the-art methods.

Framework

framework

TODO

2024.04.12 release inference code and test set
2024.04.15 update arxiv paper with camera-ready version

Dataset

We provide the test set used in our paper. We randomly select and annotate 100 samples for evaluation, including 60 real images from the Internet and 40 synthetic images. To ensure the representativeness of the evaluation, we consider the diversity of scenes and objects in the sample selection. In particular, we divide the test set into three categories: 32 images for add, 54 for change, and 14 for remove actions. All these 100 images are listed in the supplementary material. The test set can be downloaded from Google Drive

Installation

Requirements

Linux with Python ≥ 3.6
PyTorch == 1.10 and torchvision that matches the PyTorch installation. Install them together at pytorch.org to make sure of this.

Usage

Install required packages.

conda create --name zone python=3.8
conda activate zone
cd /path/to/code
pip install -r requirements.txt

Install Segment Anything (SAM):

pip install git+https://github.com/facebookresearch/segment-anything.git

or clone the repository locally and install with

git clone git@github.com:facebookresearch/segment-anything.git
cd segment-anything; pip install -e .

Download pretrained models

Download ViT-H SAM model.

cd ckpts/
wget https://dl.fbaipublicfiles.com/segment_anything/sam_vit_h_4b8939.pth

Other models can be directly loaded from HuggingFace.

Inference

python inference.py --instruction "make him a golden statue" --image_path "./data/dog.png"
python inference.py --instruction "wear her sunglasses" --image_path "./data/girl.png"

Accknowledgements

We thank Prompt-to-Prompt, InstructPix2Pix and Segment Anything for their great work and inspiration.

Citation

If you find this code/paper useful, please consider citing:

@article{li2023zone,
  title={Zone: Zero-shot instruction-guided local editing},
  author={Li, Shanglin and Zeng, Bohan and Feng, Yutang and Gao, Sicheng and Liu, Xuhui and Liu, Jiaming and Lin, Li and Tang, Xu and Hu, Yao and Liu, Jianzhuang and others},
  journal={arXiv preprint arXiv:2312.16794},
  year={2023}
}