The PCD obtained by velodyne16 does not seem to yield good voxels
af-doom opened this issue · comments
Hi @af-doom, I apologize for my late reply. Have you solved this issue? Can you share the point cloud, image, and rosbag so I can help you to debug it?
Thank you very much for your reply. The image, bag, and PCD data are attached. Thank you very much test_data.zip
…
------------------ 原始邮件 ------------------ 发件人: "hku-mars/mlcc" @.>; 发送时间: 2023年9月11日(星期一) 晚上11:54 @.>; @.@.>; 主题: Re: [hku-mars/mlcc] The PCD obtained by velodyne16 does not seem to yield good voxels (Issue #23) Hi @af-doom, I apologize for my late reply. Have you solved this issue? Can you share the point cloud, image, and rosbag so I can help you to debug it? — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you were mentioned.Message ID: @.***>
Hi @af-doom, I can't find your attached files here. Can you double-check?
Hi, how did you resolve?
I also have this problem.Did you solve it?
@Chatoyant19 tbh i wasn't able to reproduce their results with their data, what about you?
@Chatoyant19 tbh i wasn't able to reproduce their results with their data, what about you?
I can get a good result with their data.I think it is necessary to adjust some parameters, when using myself datas. Because after I changed 'voxel_size' and 'eigen_ratio', I can extract more useful planes.But it is not enough, for example, the same wall cannot be extracted completely when the plane is extracted.
@Chatoyant19 tbh i wasn't able to reproduce their results with their data, what about you?
I can get a good result with their data.I think it is necessary to adjust some parameters, when using myself datas. Because after I changed 'voxel_size' and 'eigen_ratio', I can extract more useful planes.But it is not enough, for example, the same wall cannot be extracted completely when the plane is extracted.
I think at least it is important to reproduce what they say " the average translation and rotation errors down to 6mm and 0.09 degrees for LiDAR-camera" , which for me is more or less angular error 0.067 degree baseline error 0.025 m ( should be 0.006).
With this being said I will also try to investigate more eigen and voxel size... I was working more on the edge detection intensity as my data are synthetic and highly defined...
Can I ask you which error did you introduce to your initial extrinsics?
I tried max 0.1 in translation and 2/3 degrees in rotation per axes, as with 5 degrees - as written in the paper - the algorithm goes completely wrong
@Chatoyant19 did you obtain better results?