Drag Your GAN: Interactive Point-based Manipulation on the Generative Image Manifold
Task: DragGAN
Synthesizing visual content that meets users’ needs often requires flexible and precise controllability of the pose, shape, expression, and layout of the generated objects. Existing approaches gain controllability of generative adversarial networks (GANs) via manually annotated training data or a prior 3D model, which often lack flexibility, precision, and generality. In this work, we study a powerful yet much less explored way of controlling GANs, that is, to "drag" any points of the image to precisely reach target points in a user-interactive manner, as shown in Fig.1. To achieve this, we propose DragGAN, which consists of two main components: 1) a feature-based motion supervision that drives the handle point to move towards the target position, and 2) a new point tracking approach that leverages the discriminative generator features to keep localizing the position of the handle points. Through DragGAN, anyone can deform an image with precise control over where pixels go, thus manipulating the pose, shape, expression,and layout of diverse categories such as animals, cars, humans, landscapes, etc. As these manipulations are performed on the learned generative image manifold of a GAN, they tend to produce realistic outputs even for challenging scenarios such as hallucinating occluded content and deforming shapes that consistently follow the object’s rigidity. Both qualitative and quantitative comparisons demonstrate the advantage of DragGAN over prior approaches in the tasks of image manipulation and point tracking. We also showcase the manipulation of real images through GAN inversion.
 |
 |
| Model | Dataset | Comment | FID50k | Precision50k | Recall50k | Download |
|---|---|---|---|---|---|---|
| stylegan2_lion_512x512 | Internet Lions | self-distilled StyleGAN | 0.0 | 0.0 | 0.0 | model |
| stylegan2_elphants_512x512 | Internet Elephants | self-distilled StyleGAN | 0.0 | 0.0 | 0.0 | model |
| stylegan2_cats_512x512 | Cat AFHQ | self-distilled StyleGAN | 0.0 | 0.0 | 0.0 | model |
| stylegan2_face_512x512 | FFHQ | self-distilled StyleGAN | 0.0 | 0.0 | 0.0 | model |
| stylegan2_horse_256x256 | LSUN-Horse | self-distilled StyleGAN | 0.0 | 0.0 | 0.0 | model |
| stylegan2_dogs_1024x1024 | Internet Dogs | self-distilled StyleGAN | 0.0 | 0.0 | 0.0 | model |
| stylegan2_car_512x512 | Car | transfer from official training | 0.0 | 0.0 | 0.0 | model |
| stylegan2_cat_256x256 | Cat | transfer from official training | 0.0 | 0.0 | 0.0 | model |
To run DragGAN demo, please follow these two steps:
First, put your checkpoint path in ./checkpoints, e.g. ./checkpoints/stylegan2_lions_512_pytorch_mmagic.pth. To be specific,
mkdir checkpoints
cd checkpoints
wget -O stylegan2_lions_512_pytorch_mmagic.pth https://download.openxlab.org.cn/models/qsun1/DragGAN-StyleGAN2-checkpoint/weight//StyleGAN2-Lions-internetThen, try on the script:
python demo/gradio_draggan.py@inproceedings{pan2023drag,
title={Drag your gan: Interactive point-based manipulation on the generative image manifold},
author={Pan, Xingang and Tewari, Ayush and Leimk{\"u}hler, Thomas and Liu, Lingjie and Meka, Abhimitra and Theobalt, Christian},
booktitle={ACM SIGGRAPH 2023 Conference Proceedings},
pages={1--11},
year={2023}
}