MobileNetV2 SSD FPN-Lite

모델 zoo 에서 보다보면 먼가 주르르륵 붙는데

솔찍히 mobilenet v2가 무엇인지, ssd는 또 무엇인지 몰라서 헷갈려서 조사

근데.. FPN은 전에 검색해둔 기억이 있는데 가물가물하네..

 

base network (MobileNetV2)

- 신경망(neural network)로 구성되어 있으며 분류(classification)나 탐지(detection)에 사용이 가능함

- 네트워크의 마지막에 softmax 레이어가 있으면 분류로 작동

 

detection network (Single Shot Detector or SSD)

- SSD(Single ShotDetection)나 RPN(Regional Proposal Network / R-CNN)를 이용하여

- 이미지 내의 여러 물체를 감지하고, 지역을 제안(ROI)함.

- R-CNN : Regions with Convolutional Neural Networks

[링크 : https://kr.mathworks.com/help/vision/ug/getting-started-with-r-cnn-fast-r-cnn-and-faster-r-cnn.html]

 

feature extractor (FPN-Lite)

SSD의 경우 너무 가깝거나(즉 너무 크거나, 일부만 확대되어 보일 경우) 작은경우(멀거나, 원래 작거나) 탐지를 못하는 경우가 있어

피라미드 모양으로 쌓은 FPN(Feature Pyramid Network)를 통해 특징을 추출하여 다양한 규모의 물체를 감지

 

SSD 에서는 Pyramid Feature Hierachy 라는 방식을 이용하여, 서로 다른 스케일의 특징 맵을 이용하여 멀티 스케일 특징을 추출

FPN 모델은 Region Proposeal Network (RPN) 와 Fast R-CNN을 기반으로 한다.

[링크 : https://eehoeskrap.tistory.com/300]

 

근데 내용만 봐서는 SSD + mobilenet v2 + FPN이 조합이 가능한건지 모르겠다?

 

In the MobileNetV2 SSD FPN-Lite, we have a base network (MobileNetV2), a detection network (Single Shot Detector or SSD) and a feature extractor (FPN-Lite). Base network: MobileNet, like VGG-Net, LeNet, AlexNet, and all others, are based on neural networks. The base network provides high-level features for classification or detection. If you use a fully connected layer and a softmax layer at the end of these networks, you have a classification.   Example of a network composed of many convolutional layers. Filters are applied to each training image at different resolutions, and the output of each convolved image is used as input to the next layer (source Mathworks) But you can remove the fully connected and the softmax layers, and replace it with detection networks, like SSD, Faster R-CNN, and others to perform object detection. Detection network: The most common detection networks are SSD (Single Shot Detection) and RPN (Regional Proposal Network). When using SSD, we only need to take one single shot to detect multiple objects within the image. On the other hand, regional proposal networks (RPN) based approaches, such as R-CNN series, need two shots, one for generating region proposals, one for detecting the object of each proposal. As a consequence, SSD is much faster compared with RPN-based approaches but often trades accuracy with real-time processing speed. They also tend to have issues in detecting objects that are too close or too small. Feature Pyramid Network: Detecting objects in different scales is challenging in particular for small objects. Feature Pyramid Network (FPN) is a feature extractor designed with feature pyramid concept to improve accuracy and speed.

[링크 : https://docs.edgeimpulse.com/docs/edge-impulse-studio/learning-blocks/object-detection/mobilenetv2-ssd-fpn]