Path Aggregation Network (PANet)

Mask R-CNN + FPN + enhancing information propagation

Mask R-CNN과의 차이점

 

1. 피쳐 추출단에서 - 새롭게 증강된 방식의 bottom-up pathway FPN 구조를 이용한다.

 

이렇게 하면 low-layer의 특징들이 좀 더 잘 전파된다고 한다.

 

 2. adaptative feature pooling를 이용하여 모든 레벨의 피쳐에서 제안들을 다 추출한다.

 

3. 예측된 픽셀의 위치 정합성을 향상시키위해 FCN의 conv layer의 출력 결과와 fully connected layer를 이어주는 통로를 만들었다. added a path processing the output of a convolutional layer of the FCN with a fully connected layer to improve the localisation of the predicted pixels.

Lateral connection between the top-down pathway and the augmented bottom-up pathway. Source:  S. Liu et al. (2018)

Adatative feature pooling layer. Source:  S. Liu et al. (2018)

Branch of the PANet predicting the binary mask using a FCN and a new path with a fully connected layer. Source:  https://arxiv.org/pdf/1803.01534.pdf

PANet Achitecture. (a): Feature extractor using the FPN achitecture. (b): The new augmented bottom-up pathway added to the FPN architecture. ©: The adaptative feature pooling layer. (d): The two branches predicting the bounding box coordinated and the object class. (e): The branch predicting the binary mask of the object. The dashed lines correspond to links between low-level and high level patterns, the red one is in the FPN and consists in more than 100 layers, the green one is a shortcut in the PANet consisting of less than 10 layers. Source:  S. Liu et al. (2018)

 

성능

 ResNeXt 를 특징 추출기로 써서 2016 COCO segmenation challenge에서 42.0% AP score를 달성

2017 COCO segmentation challenge에선 46.7% AP 달성 (using a ensemble of seven feature extractors: ResNet (K. He et al. (2015), ResNeXt (S. Xie et al. (2016)) and SENet (J. Hu et al.(2017)).)

 

ref) https://medium.com/@arthur_ouaknine/review-of-deep-learning-algorithms-for-image-semantic-segmentation-509a600f7b57

Review of Deep Learning Algorithms for Image Semantic Segmentation