search

WeijingShi / Point-GNN

Point-GNN: Graph Neural Network for 3D Object Detection in a Point Cloud, CVPR 2020.
MIT License
523 stars 114 forks source link

点云检测结果有误 #63

Open EvaneSnow opened 3 years ago

EvaneSnow commented 3 years ago

@WeijingShi 您好，再次麻烦您。我使用自己的128线雷达点云数据进行预测，发现点云结果很乱，点云没有检测到！不像KITTI数据集展示的效果很规矩，不知出现在什么原因，希望作者及时回复。

EvaneSnow commented 3 years ago

@WeijingShi 您好，作者！run.py里面是不是对输入的bin文件的点云点进行了筛选？不知是什么原因，之前用的KITTI数据集进行检测，效果很好，换成自己录制的128线数据之后，点云检测一个框也没有，显示出来的open3d点云的点比之前播放bag的时候少很多，希望作者及时回复，谢谢！

WeijingShi commented 3 years ago

Hi @EvaneSnow, yes, we only keep the points that are visible in the front image: https://github.com/WeijingShi/Point-GNN/blob/48f3d79d5b101d3a4b8439ba74c92fcad4f7cab0/dataset/kitti_dataset.py#L643

If your 128-line data and calibration files are similar to Kitti, you have probably seen around a 70-degree view of the point cloud.

As for the detection results, I am sure about how large is the gap between your custom data and kitti. If you have labels, re-training may be an option.

bingo9062 commented 2 years ago

@EvaneSnow 请问您的问题解决了吗？我也遇到了128线自己的数据集，识别不出，方便联系吗？非常感谢