M2Det : Highly Accurate Object Detection Model

This is an introduction to「M2Det」, a machine learning model that can be used with ailia SDK. You can easily use this model to create AI applications using ailia SDK as well as many other ready-to-use ailia MODELS.

Overview

M2Det is a highly accurate object detection model proposed in November 2018. It can detect bounding boxes of objects from the 80 categories in COCO.

M2Det: A Single-Shot Object Detector based on Multi-Level Feature Pyramid NetworkFeature pyramids are widely exploited by both the state-of-the-art one-stage object detectors (e.g., DSSD, RetinaNet…arxiv.org

Conventional object detection uses an image classification backbone (Mobilenet, VGG, ResNet, etc.) to compute bounding boxes. For example, in SSD, after extracting features with the image classification backbone, the bounding box is calculated by adding a similar backbone in the later stage.

The same was true for ReinaDet, which uses Feature Pyramids, and although it uses more hierarchical features than SSD, the backbone of object detection was based on image classification.

In M2Det, after extracting features with the image classification backbone, a specialized backbone is used for object detection to achieve higher accuracy.

（Source：https://arxiv.org/abs/1811.04533）

This backbone specialized for object detection is called MLFPN (Multi-Level Feature Pyramid Network), which uses TUM (Thinned U-shape Module) in a hierarchical manner.