chin-editing-dataset is a high-quality image dataset of double chin editing, based on StyleGAN2.
The dataset is built by our diffusion method(See the Section3.4 in our paper)
Coarse-to-Fine: Facial Structure Editing of Portrait Images via Latent Space Classifications
Yiqian Wu, Yongliang Yang, Qinjie Xiao,Xiaogang Jin*.
ACM Transactions on Graphics (Proc. of Siggraph'2021), 2021, 40(4): Article 46.
[demo]
[code] coming soon.
We create the first large-scale chin editing dataset to facilitate future research. The dataset contains 14,788 pairs of realistic portrait images at 1024×1024 resolution with and without a double chin and their corresponding latent codes.
All the images are synthetic and generated by StyleGAN2.
google drive link of the dataset: https://drive.google.com/drive/folders/10e6WB4YLb3Mn6Us4mPAgBksGr7kBx8q0?usp=sharing
| dir | information |
|---|---|
| ├ double_chin_pair_psi_0.5 | data for truncation_psi-0.5 |
| │ ├ codes | latent codes. {img_id}_wp.npy : the original latent code, {img_id}_inverted_WP_codes.npy : the latent code after removing double chin. |
| │ ├ diffused | the images that generated directly from {img_id}_inverted_WP_codes.npy |
| │ ├ res | results images , {img_id}.jpg |
| │ └ origin | origin images, {img_id}.jpg |
| ├ double_chin_pair_psi_0.8 | data for truncation_psi-0.8 |
| │ ├ codes | latent codes. {img_id}_wp.npy : the original latent code, {img_id}_inverted_wp.npy : the latent code after removing double chin. |
| │ ├ res | results images, {img_id}.jpg |
| │ └ origin | origin images, {img_id}.jpg |
Analyzing and Improving the Image Quality of StyleGAN
Tero Karras, Samuli Laine, Miika Aittala, Janne Hellsten, Jaakko Lehtinen, Timo Aila
- The chin-editing-dataset is available for non-commercial research purposes only.
coming soon.