ProFill: High-Resolution Image Inpainting with Iterative Confidence Feedback and Guided Upsampling, ECCV 2020
Yu Zeng$^1$, Zhe Lin$^2$, Jimei Yang$^2$, Jianming Zhang$^2$, Eli Shechtman$^2$, Huchuan Lu$^1$
1 Dalian University of Technology,
2 Adobe Research
| Results | Web App | REST API | Paper | Supplementary Material | Comparison | Object masks |
This API is for research purposes only and cannot be used for commertial purposes in any form
High-resolution results (≥1024)
Synthetic samples | Removing real objects
Synthetic samples
Removing real objects
Web APP
Usage:
Effect of specifying the include or avoid region:
API
http://47.57.135.203:2333/api
This API is for research purposes only and cannot be used for commertial purposes in any form
mode_img = img.mode
mode_msk = mask.mode
W, H = img.size
str_img = img.tobytes().decode("latin1")
str_msk = mask.tobytes().decode("latin1")
data = {'str_img': str_img, 'str_msk': str_msk, 'width':W, 'height':H,
'mode_img':mode_img, 'mode_msk':mode_msk}
# Enable upsample
#data = {'str_img': str_img, 'str_msk': str_msk, 'str_include':None, 'str_avoid':str_avoid, 'width':W, 'height':H, #'mode_img':mode_img, 'mode_msk':mode_msk, 'is_refine': True}
r = requests.post('http://47.57.135.203:2333/api', json=data)
Object masks
wget https://maildluteducn-my.sharepoint.com/:u:/g/personal/zengyu_mail_dlut_edu_cn/EUcTzYoX5LhFkBCjG62E5wwBzru1eM4PhwmRNGmo08pH5Q?download=1
mv EUcTzYoX5LhFkBCjG62E5wwBzru1eM4PhwmRNGmo08pH5Q?download=1 object_masks.zip
unzip object_masks.zip
Supplementary Material
Comparison
- Visual comparison of ours (ProFill and +Upsample) with HiFill
- Visual comparison with state-of-the-art methods
- Quantitative comparison with more methods