clynamen / nuscenes2bag Goto Github PK

Thanks for making this useful tool. Unfortunately I am stuck using an older version of ROS.

I forked the code and modified it to compile on Ubuntu 16.04 or 18.04 using only the standard version of libraries that came with that distribution. I also added an alternative implementation of a thread pool, and the option to publish the 5th data field "ring" from the laser sensor.

Ubuntu 16.04 (ROS Kinetic Kame) Branch: ubuntu_1604
Ubuntu 18.04 (ROS Melodic Morenia) Branch: ubuntu_1604
Ubuntu 19+ Branch: master

I tested it on those two distributions only. I don't have any other linux to try.

The backported code uses precompiler logic that switches based on the detected library versions.
For example:

#if CMAKE_CXX_STANDARD >= 17
  std::optional<SceneInfo> getSceneInfo()
#else
  boost::shared_ptr<SceneInfo> getSceneInfo()
#endif

which means that the code is more messy, but this saves trying to maintain two very different branches.

The code in NuScenes2Bag.cpp uses a thread pool from Boost 1.66 so I found another implementation.

@clynamen @ChernoA I played the ROS bag file in loop with robag play -l 61.bag and then rotopic echo /tf. Please find below for camera_front/raw.

I wanted to read the camera intrinsic parameters as well in a program, but I unable to find this data in the in the /tf topic. Hence, I was wondering if the camera intrinsic parameters defined in the calibrated_sensor.json in the nuScenes dataset could also be published in the /tf topic?

What new features are people interested in?
e.g.

• Lidar rings
• Radar points
• Bounding boxes

Comment here with what features you want to see, and if there's enough interest then maybe we can add them.

Sorry for using this 'issues' board for discussion ;-)

Hey, nice work for writing this converting tool. It works great!!!

Would it be possible to add the Camera info into the rosbag file? When I subscribe to the "/cam_front/raw" topic. I only find the Images. Sure I can get the camera info somewhere else. However, if it's not too difficult, it would be great if you can add that!

Thanks!

Hello,
Thanks for making this great tool. It is very useful!

I want to suggest some minor changes to the code, like to specify the C++17 version and the correct minimum versions of CMake and Boost.
Here is a Pull Request #2
Your comments are welcome.

Hi,

Thank you for all your work on this package. I used nuscenes2bag to convert nuscenes-mini and it went from being 5 GB to around 62 GB. Is this normal and if so, would you mind explaining why the output is so much larger?

Thanks!
Eric

Hi,
Is it possible to convert the nuscenes annotations to /marker/annotations topic as well?

Thanks

Hi, I try to built on ubuntu18.04 with melodic

but I can got some errors:

  Could not find a package configuration file provided by "rosbag" with any

I try to install sudo apt install ros-melodic-rosbag, it says this package can not found on ros server

In the above image radar data(.pcd) is in RadarObjects datatype, lidar data(.pcd) is in sensor_msgs/PointCloud2 , how to convert RadarObject to PointCloud2 datatype format?

@clynamen I was wondering how can we inference some of the pre-trained models in Det3D and draw the bounding boxes over camera images (front, front right, front left, back, back right and back left) and also over the point cloud lidar_top data plotted with rviz?