Tom-Huang / carla_data_collection

set sensors on an autonomous agents(a vehicle) in carla and collect albedo, depth, reflection vector, semantic data and rgb image
20 stars 4 forks source link

How to generate lidar point cloud? #4

Open Sibozhu opened 4 years ago

Sibozhu commented 4 years ago

Dear @Tom-Huang ,

Thanks for the previous help in compiling your code.

In the README, you said You can modify the code to generate the lidar point cloud, can you actually point out which of those lines to be uncomment? I found there's a function called lidar_saver but it isn't being called anywhere.

Once I can save and generate the correct lidar point cloud with semantic label, I'll clean up the code and submit a pull request to you, so people in the future can benefit as well.

Best, Sibo

Sibozhu commented 4 years ago

Also, I found code for BufferedImageSaver from somewhere else, but couldn't find BufferedLidarSaver, is this a necessary function for generating lidar point cloud?

maskjp commented 4 years ago

Hi, @Sibozhu! Can I ask a question about the recompile? Except for copy the uasset files in the folder, do we need to change anything in the project file for recompiling Carla? Because after I recompile Carla, I still can't use the normal camera in the file. And I opened the unreal project file. I found the project didn't load the new uasset file.

Thank you!

Sibozhu commented 4 years ago

@maskjp ,have you tried issue #2 ?

maskjp commented 4 years ago

@maskjp ,have you tried issue #2 ?

Hi, @Sibozhu! Thank you for your reply. Yes, I deleted the build folder and run the make clean command, but the project still can't load the normal camera and others. I only want to use the lidar semantic so I removed code related, to these sensors. But I found the code only can save the image. And no code saves the lidar data.

Btw, saving images seems not to require BufferedImageSaver.

Sibozhu commented 4 years ago

@maskjp You are absolutely right, using BufferedImageSaver will save the image into .ply format, which is the ideal format that we want to save lidar point cloud into. I believe the BufferedLiDARSaver is code modified from BufferedImageSave.

project can't load normal cameras meaning you still can't save images?

If you are looking for a codebase that compiles and can generate rgb and depth image (semantic segmentation image still need to be post-processed, which we can make it generated in real-time in the future), you can try my fork. We can work on making the codebase clean and functioning together and submit pull request to @Tom-Huang in the end. (also added autopilot feature so the simulated agent can have some driving footage)

demo

maskjp commented 4 years ago

@maskjp You are absolutely right, using BufferedImageSaver will save the image into .ply format, which is the ideal format that we want to save lidar point cloud into. I believe the BufferedLiDARSaver is code modified from BufferedImageSave.

project can't load normal cameras meaning you still can't save images?

If you are looking for a codebase that compiles and can generate rgb and depth image (semantic segmentation image still need to be post-processed, which we can make it generated in real-time in the future), you can try my fork. We can work on making the codebase clean and functioning together and submit pull request to @Tom-Huang in the end. (also added autopilot feature so the simulated agent can have some driving footage)

demo

@Sibozhu Thank you for your reply! I mean, the Unreal engine project can't load newly added uassets even after I deleted all the intermediate files and recompile the Carla. But I am currently interested in getting labeled lidar data, so I may don't need the four new uassets.

BTW, to save lidar point cloud, I found a sample code in carla and the save_to_disk method.

Sibozhu commented 4 years ago

@maskjp You are absolutely right, using BufferedImageSaver will save the image into .ply format, which is the ideal format that we want to save lidar point cloud into. I believe the BufferedLiDARSaver is code modified from BufferedImageSave. project can't load normal cameras meaning you still can't save images? If you are looking for a codebase that compiles and can generate rgb and depth image (semantic segmentation image still need to be post-processed, which we can make it generated in real-time in the future), you can try my fork. We can work on making the codebase clean and functioning together and submit pull request to @Tom-Huang in the end. (also added autopilot feature so the simulated agent can have some driving footage) demo

@Sibozhu Thank you for your reply! I mean, the Unreal engine project can't load newly added uassets even after I deleted all the intermediate files and recompile the Carla. But I am currently interested in getting labeled lidar data, so I may don't need the four new uassets.

BTW, to save lidar point cloud, I found a sample code in carla and the save_to_disk method.

Thanks! @maskjp This is definitely helpful!