[DL Hacks]Smart, Sparse Contours to Represent and Edit Images
0 likes652 views
This document summarizes a CVPR2018 paper about representing and editing images using sparse contours. It encodes images using sparse contours and additional gradient or color channels, which allows the image to be reconstructed from the code and also edited by modifying the code. The encoding represents the image in a way that is easy for humans to understand and edit, and also easy for machines to reconstruct the original image. An online demo is available that demonstrates reconstruction and editing capabilities.
![Info
• Sparse, Smart Contours to Represent and
Edit Image [CVPR2018]
- https://contour2im.github.io/
- https://arxiv.org/abs/1712.08232
3](https://image.slidesharecdn.com/201872dlhacks-180706002502/85/DL-Hacks-Smart-Sparse-Contours-to-Represent-and-Edit-Images-4-320.jpg)
![Encoder Overview
5
1. Extract Edge [1]
2. Get Contour
(remove less than
10pixel-connected)
3. Compute addiConal
channel
[1] Structured forests for fast edge detecCon [Dollar, ICCV2013]](https://image.slidesharecdn.com/201872dlhacks-180706002502/85/DL-Hacks-Smart-Sparse-Contours-to-Represent-and-Edit-Images-6-320.jpg)
[DL Hacks]Smart, Sparse Contours to Represent and Edit Images
- 1. 【CVPR2018論文紹介 】 Sparse, Smart Contours to Represent and Edit Images 2018/07/02 DLHacks Otsubo
- 2. Topic1 : Image Encoding and Decoding (Code = Sparse Contour + α) 1
- 3. Topic2 : Image Editing 2
- 4. Info • Sparse, Smart Contours to Represent and Edit Image [CVPR2018] - https://contour2im.github.io/ - https://arxiv.org/abs/1712.08232 3
- 5. 選んだ理由 • Contourからの画像復元とか無理では？ あまりにもill-posed → 少量の情報（勾配）を追加するだけで実現 • Encoding Decodingにおいて 人間が理解，編集しやすく 機械が復元しやすいCodeを模索 • イケメンになりたい 4
- 6. Encoder Overview 5 1. Extract Edge [1] 2. Get Contour (remove less than 10pixel-connected) 3. Compute addiConal channel [1] Structured forests for fast edge detecCon [Dollar, ICCV2013]
- 7. Encoder Additional Channel 1. Color - (Rleft, Gleft, Bleft, Rright, Gright, Bright ) 2. Gradient 編集に最適(reconstructionでは弱い) - (Rdx, Gdx, Bdx, Rdy, Gdy, Bdy ) 3. Learned Feature - Semantic Segmentationのように学習 6
- 14. Editing 13