Skip to content

Commit

Permalink
update
Browse files Browse the repository at this point in the history
  • Loading branch information
yucornetto committed Dec 15, 2020
1 parent 1ee451e commit cbdca81
Show file tree
Hide file tree
Showing 3 changed files with 30 additions and 24 deletions.
10 changes: 10 additions & 0 deletions result/IMAGE.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,10 @@
## Image Matting
<p align="center">
<img src="1.png" width="1050" title="Matting Comparison"/>
</p>
<p align="center">
<img src="2.png" width="1050" title="Matting Comparison"/>
</p>
<p align="center">
<img src="3.png" width="1050" title="Matting Comparison"/>
</p>
26 changes: 2 additions & 24 deletions result/RESULT.md
Original file line number Diff line number Diff line change
Expand Up @@ -2,34 +2,12 @@
We provide visual samples and comparison with Internet Images here.

## Image Matting
<p align="center">
<img src="1.png" width="1050" title="Matting Comparison"/>
</p>
<p align="center">
<img src="2.png" width="1050" title="Matting Comparison"/>
</p>
<p align="center">
<img src="3.png" width="1050" title="Matting Comparison"/>
</p>
Please refer to [IMAGE.md](IMAGE.md) for details.

## Fully Automatic Matting System
We further train MG Matting with an internal portrait matting dataset consisting of 4395 samples, and combine it with a base segmentation model to obtain a fully automatic matting system. We compare this system to latest trimap-free matting system (MODNet)[https://github.com/ZHKKKe/MODNet] and also commercial software Photoshop 2021.

<p align="center">
<img src="4.png" width="1050" title="System Comparison"/>
</p>
<p align="center">
<img src="5.png" width="1050" title="System Comparison"/>
</p>
<p align="center">
<img src="6.png" width="1050" title="System Comparison"/>
</p>
<p align="center">
<img src="7.png" width="1050" title="System Comparison"/>
</p>
<p align="center">
<img src="8.png" width="1050" title="System Comparison"/>
</p>
Please refer to [SYSTEM.md](SYSTEM.md) for details.

## Video Matting
We note that MG Matting, though not utilizing temporal information yet, can potentially produce great results on videos. Please refer to the following links for video demos.
Expand Down
18 changes: 18 additions & 0 deletions result/SYSTEM.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,18 @@
## Fully Automatic Matting System
We further train MG Matting with an internal portrait matting dataset consisting of 4395 samples, and combine it with a base segmentation model to obtain a fully automatic matting system. We compare this system to latest trimap-free matting system (MODNet)[https://github.com/ZHKKKe/MODNet] and also commercial software Photoshop 2021.

<p align="center">
<img src="4.png" width="1050" title="System Comparison"/>
</p>
<p align="center">
<img src="5.png" width="1050" title="System Comparison"/>
</p>
<p align="center">
<img src="6.png" width="1050" title="System Comparison"/>
</p>
<p align="center">
<img src="7.png" width="1050" title="System Comparison"/>
</p>
<p align="center">
<img src="8.png" width="1050" title="System Comparison"/>
</p>

0 comments on commit cbdca81

Please sign in to comment.