demo.mp4
Official implementation ofFreeDrag: Feature Dragging for Reliable Point-based Image Editing.
- Authors:Pengyang Ling*,Lin Chen*,Pan Zhang,Huaian Chen, Yi Jin, Jinjin Zheng,
- Institutes:University of Science and Technology of China; Shanghai AI Laboratory
- [Paper][Project Page][Web Demo]
This repo proposes FreeDrag, a novel interactive point-based image editing framework free of the laborious and unstable point tracking process🔥🔥🔥.
To serve the intricate and varied demands of image editing, precise and flexible manipulation in image content is indispensable. Recently, Drag-based editing methods have gained impressive performance. However, these methods predominantly center on point dragging, resulting in two noteworthy drawbacks, namely "miss tracking", where difficulties arise in accurately tracking the predetermined handle points, and "ambiguous tracking", where tracked points are potentially positioned in wrong regions that closely resemble the handle points. To address the above issues, we proposeFreeDrag,a feature dragging methodology designed to free the burden on point tracking. TheFreeDragincorporates two key designs, i.e., template feature via adaptive updating and line search with backtracking, the former improves the stability against drastic content change by elaborately controls feature updating scale after each dragging, while the latter alleviates the misguidance from similar points by actively restricting the search area in a line. These two technologies together contribute to a more stable semantic dragging with higher efficiency. Comprehensive experimental results substantiate that our approach significantly outperforms pre-existing methodologies, offering reliable point-based editing even in various complex scenarios.
[2024/03/06]FreeDragis accepted by CVPR 2024.
[2023/12/11] The updataedFreeDragcontaining the implementations in both StyleGAN and Diffusion models is available now.
[2023/12/8] FreeDrag based on diffusion model is available now, which support dragging editing in both real images and generated images.
[2023/7/31] The web demo (StyleGAN) inOpenXLabis available now.
[2023/7/28] The function of real image editing is available now.
[2023/7/15] Code of local demo is available now!💥
[2023/7/11] Thepaperandproject pageare released!
- Local demo of FreeDrag
- Web demo of FreeDrag
- Diffusion-based FreeDrag
First clone our repository
git clone --depth=1 https://github /LPengYang/FreeDrag
To create a new environment, please follow the requirements ofNVlabs/stylegan2-ada.
Notice:It is observed that the errors (setting up PyTorch plugin “bias_act_plugin “... Failed or “upfirdn2d_plugin “... Failed) may appear in some devices, we hope these potential solutions (1,2,3,4) could be helpful in this case.
Then install the additional requirements
pip install -r requirements.txt
Then download the pre-trained models of stylegan2
bash download_models.sh
Notice:The first model (face model) could be downloaded very slowly in some cases. In this case, it is suggested to restart the download (works sometimes) or directly download it from thislink,please download the correct model (ffhq-512×512) and renamed it as "faces.pkl" and manually put it in the created checkpoints file (after all the other models are downloaded).
Finally initialize the gradio platform for interactive point-based manipulation
CUDA_LAUNCH_BLOCKING=1 Python FreeDrag_gradio.py
You can also upload your images and then edit them. For a high-quality image inversion, it is suggested to make sure that the resolution and style (such as layout) of the uploaded images are consistent with the generated images of corresponding model. The resolution of different model is listed as follows:
| Model | face | horse | elephant | lion | dog | bicycle | giraffe | cat | car | church | metface |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Resolution | 512 | 256 | 512 | 512 | 1024 | 256 | 512 | 512 | 512 | 256 | 1024 |
The proposedFreeDragBench Datasetis available on thewebsite.
All codes used or modified fromStyleGAN2are under theNvidia Source Code License. The code related to the FreeDrag algorithm is only allowed for personal activity. The diffusion-based FreeDrag is implemented based onDragDiffusion.
If you find our work helpful for your research, please consider citing the following BibTeX entry.
@article{ling2023freedrag,
title={Freedrag: Point tracking is not you need for interactive point-based image editing},
author={Ling, Pengyang and Chen, Lin and Zhang, Pan and Chen, Huaian and Jin, Yi},
journal={arXiv preprint arXiv:2307.04684},
year={2023}
}