0
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?

More than 5 years have passed since last update.

【Stereo Depth】AnyNet : 複数解像度でStereo Matchingを行う

Posted at

Anytime Stereo Image Depth Estimation on Mobile Devices

![image.png](https://qiita-image-store.s3.ap-northeast-1.amazonaws.com/0/482094/e87e78b8-3ae4-82f7-4a68-24a18f77b049.png)

2 stage Refinementならず4 stageのネットワークを作ったら良い感じだったらしい。

新規性

![image.png](https://qiita-image-store.s3.ap-northeast-1.amazonaws.com/0/482094/3b855bdf-db55-23c5-f704-07cc0b17014f.png)

U-Net Feature Extractor

![image.png](https://qiita-image-store.s3.ap-northeast-1.amazonaws.com/0/482094/2439a215-8eb1-c9a9-788e-16cc28062c04.png) Unetを使うことで同時に複数サイズのFeature Mapを得れるのね。賢い

Distance-Based Cost Volume

![image.png](https://qiita-image-store.s3.ap-northeast-1.amazonaws.com/0/482094/5ddb727a-3b40-96ee-0e3f-5cb2c76b821e.png) Cost Volumeを作る際はPSMneのようにConcatenationするか、Inner Productを取る方法が一般的であるが、Anynetでは差分を取っている。 個人的にInner Productの方が相関性が見えるから良いと思ったのだが。

恐らくInner Productをやっていないから、実験したら面白いかも。
計算コストは同じはずだし

結果

![image.png](https://qiita-image-store.s3.ap-northeast-1.amazonaws.com/0/482094/ba681f86-c1cc-f960-9de6-2eb0e9b49405.png) かなり早い!

image.png
PSMNetに比べて精度は劣っているがスピードが格段に早い

結論

・Real time systemに使いたいならAnyNetあり Stereo Depthで初の精度のあるReal time system(10Hz)なのではないだろうか

参考文献

Anytime Stereo Image Depth Estimation on Mobile Devices https://arxiv.org/pdf/1810.11408.pdf
0
0
0

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do with signing up
0
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?