zhengjun-du / lc-crf-slam Goto Github PK

i have some questions about the third party dense crf. it is a wonderful module that i need, i wonder if there are some examples about how to use it on point cloud data such as kitti velodyne to do semantic segmentation, another question is how to use the gpu version, the densecrf_gpu.cuh and pairwise_gpu.cuh cannot be found, i dont know what to do. thanks a lot

One more question, is the ORB window running with dynamic feature points removed? In the other window, accurate label, are the red dots dynamic points that are displayed in real time?

运行/Examples/RGB-D/rgbd_tum /home/ch/data/rgbd_dataset_freiburg3_walking_xyz/ /home/ch/data/associations.txt Examples/RGB-D/TUM3.yaml

Loading ORB Vocabulary. This could take a while...
已杀死

I have successfully compiled your work, but when I run it, it fails with Segmentation fault after "loading ORB Vocabulary".
Do you have any idea about what should I do or modify? Many thanks!
Ubuntu16+Opencv3.4.3+Eigen3.1.0

请问论文里面提到的“长期”是不是指利用了相邻多个关键帧的意思呢？而其他论文是“短期”的，只是利用了相邻的普通帧。我的理解正确吗？
然后，LC-CRF是针对特征点处理的，如果要对动态区域处理（比如在重建稠密地图时），需要利用聚类或者图像分割算法是吧？

I would like to ask whether CRF can be combined with the target detection network to achieve dynamic point detection and elimination

您好，我正在看您的文章《LC-CRF-SLAM》，关于论文中报道的位姿估计精度，我有一些疑问。
在TABLE2中，Bonn数据集“person_tracking”上的t.RPE部分给出的Mean是1.193，OVD-SLAM中对LC-CRF-SLAM这一项的报道是0.396，而我自己测出来的是0.042左右。我想这是不是因为使用evo的时候使用的标准不同导致的。
我在测t.RPE的时候使用代码是“evo_rpe tum groundtruth.txt LCCRF.txt -vpa”，请问这是否和您使用的一致？
还有一个疑问，在TUM数据集中，t.RPE都是0.0X量级的，为什么在Bonn中会是0.X量级呢？
谢谢！

I have a question if LC-CRF can work without GCRANSAC?