0
1

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 3 years have passed since last update.

【Panoptic Segmentation】Semantic FPN : FPNはなかなか効率的らしい

Last updated at Posted at 2020-12-04

Panoptic Feature Pyramid Networks

![image.png](https://qiita-image-store.s3.ap-northeast-1.amazonaws.com/0/482094/a151327e-0e75-8e84-37fb-2c31d5dc78c9.png)

Panoptic SegmentationにFPN(Feature Pyramid Network)を使用してみたという論文である

新規性

![image.png](https://qiita-image-store.s3.ap-northeast-1.amazonaws.com/0/482094/47fcaab8-2d1d-5835-29bb-3b6d3f67f766.png)

PanopticでFPNを使ってみた

![image.png](https://qiita-image-store.s3.ap-northeast-1.amazonaws.com/0/482094/1f1dc613-449c-8b55-5fb7-838412f036a6.png) 3解像度の出力それぞれから得たROI(物体がありそうなエリア)すべてに対してSemantic Segmentationをする事でInstance Segmentationを行っている。

image.png

3解像度の特徴量を全部足したFeature Map(黒いやつ)を使ってSemantic Segmentationを行う

FPNの効率性の実験

![image.png](https://qiita-image-store.s3.ap-northeast-1.amazonaws.com/0/482094/bfd52510-e9a6-88bb-ebae-b4ee688d0a2d.png)

(b)(c)(d)が1/8の解像度で出力する時に良い結果を出せるモデルの代表例。

image.png
掛け算足し算の数(左図)とメモリの消費量(右図)を比べてみるとFPNが一番計算量が小さくてメモリの使用量が少ないのが分かる。

結論

・Dilated / Symmetric Decoder よりFPNが一番計算コストとメモリコストが小さくお得

いろんな論文にFPNが使われているのも頷ける

参考文献

Panoptic Feature Pyramid Networks https://arxiv.org/pdf/1901.02446.pdf
0
1
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
1

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?