silencht / SG-SLAM

SG-SLAM: A Real-Time RGB-D Visual SLAM toward Dynamic Scenes with Semantic and Geometric Information
GNU General Public License v3.0
157 stars 20 forks source link

What are target detection? #5

Closed 40526136 closed 1 year ago

40526136 commented 1 year ago

hello author Can I ask what objects can be detected

silencht commented 1 year ago

Since the mobilenetV3-SSD model that has been trained is used, there are 20 types of objects that can be detected, which are : object class that can be detected

The model above comes from : https://github.com/Tencent/ncnn/blob/master/examples/mobilenetv3ssdlite.cpp

You can also use other examples of ncnn or train the network model yourself.

40526136 commented 1 year ago

hello author After detecting the screen and chair objects, and in the semantic segmentation, the red and blue three-dimensional boxes are displayed on rviz, but the people and potted plants are not displayed on rviz. Where can these programs be written.

silencht commented 1 year ago

Most of the code is located in Detector2D.cc, Detector3D.cc and PointcloudMapping.cc in src/sg-slam/src. It should be noted that, as a highly dynamic object, people have been eliminated in advance when building the map.