pyslam VS ORB_SLAM2

From what I have understood after reading research papers related to the VSLAM, the modularity aspect is not easy to achieve given the extracted features and descriptors are intrinsically linked with feature matching and handling of map points. I would like to know if there are some good Open Source VSLAM projects available which can be used with different feature extractors so I can get a comparative results with respect to just changing the feature extractors . I have tried pyslam project which is actually quite good considering the modularity but as the author himself points out this is only for academic purposes and when I compared the results of ORB_SLAM2 feature extractor using this module vs the original ORB_SLAM2 for KITTI data set , the results are not comparable. I am also looking into OpenVINS ( and from initial reading it is also using ORB Features, although it does have a base Tracker class which can be modified to create a new Tracker with different descriptor) If anyone has worked with custom feature extractor incorporated into prebuilt SLAM pipeline and can guide me as to how to proceed with the implementation of custom Feature extractor into a SLAM Front end using a Open Source VSLAM framework, it will be really helpful.

I've stumbled upon this and that using DL, and will try to check to simultaneously evaluate them next to developing something using pySLAM. At least that's the current plan.

My question however, is what about certain packages or libraries that I find on github that have been developed and tested on some specific Ubuntu version? Like this pySLAM library - it was developed and tested under Ubuntu 18.04, but is it actually possible to run this on an Arch Linux machine?

ORB_SLAM2 on GitHub: https://github.com/raulmur/ORB_SLAM2

Hi everyone, I am trying to implement a VSLAM with DNN specifically the Feature Extraction module in the SLAM pipeline. Something on the lines of this repo Superpoint_SLAM , which integrates SuperPoint Feature extraction into ORB_SLAM2

Maybe you could give Orb-SLAM2 a try with a Pi cam. The package says it works with a monocular camera. https://github.com/raulmur/ORB_SLAM2

I am thinking about migrating ORBSLAM2(written in cpp) to python3. Currently there are only python3 wrappers for ORBSLAM2, like ORB_SLAM2-PythonBindings and pyORBSLAM2. The biggest problem is that I can't easily and quickly improve, change all the stuff in cpp, as it would be in python. I am aware that the cpp code is much faster the python equivalent, but implementing and improving additional features is much easier in python, at least for me. For converting the code, I'll keep the optimization stuff like g2o(g2opy), pangolin(pypangolin) and DBoW2(pyDBoW) in cpp, but the rest should be in pure python code.

This repo has an advanced SLAM implementation. They also link to a few research papers about how this group made this SLAM algorithm. The ORB SLAM 2 paper has all of the parts of modern SLAM systems, so I'd read that, and the other papers if you have time at the end.

Example: https://github.com/raulmur/ORB_SLAM2/blob/f2e6f51cdc8d067655d90a78c06261378e07e8f3/Examples/ROS/ORB_SLAM2/src/ros_stereo.cc#L71

I'm not sure what your final application is, but you can use a SLAM algorithm to recover your camera trajectory. I've used ORB SLAM 2, and it's pretty nice.