刘国庆, Guoqing Liu
刘国庆, Guoqing Liu
我觉得x1 x2的确是归一化平面上的坐标.像素坐标为p1,相机内参数矩阵为K,那么根据P86页式5.7和式7.3，有 p1=Kx1 (根据式5.7)=>x1=Pc/s1 上式中Pc为空间点在相机坐标系下的坐标，s1即为Z——Pc的z轴坐标。从上面的这个式子来看，的确x1是像素点在归一化平面上的坐标。 最初我看到这个issue的时候也在纳闷为什么x1 x2符合坐标轴变换关系。不过按照我的推导，式7.4的写法不严谨——在数学意义上等式左右两侧并不相等，可以见issue#100。
Maybe you need to compile with the correct version like 2.0.0 of Ceres-solver.
> > Maybe you need to compile with the correct version like 2.0.0 of Ceres-solver. > > I have installed Ceres2.0.0 when build cartographer, but it was in the workspace...
The files with the `.m` extension are MATLAB scripts. I think you need MATLAB to execute them.
I have meet the same problem too…… and I found some some checks on whether a point is really in the image were ignored.
> @DreamWaterFound hi, have you solved this problem? same problem here. @XinnWang Yes. Notice the function `Coco.__getitem__.points_to_2D` in file `datasets/Coco.py` ([here](https://github.com/eric-yyjau/pytorch-superpoint/blob/4ff74df8fa3c10ce9eb9fdc561f787d9e8bc9691/datasets/Coco.py#L231)), especially for COCO datasets with different image sizes, converted...
需要在[这里](https://github.com/Ewenwan/MVision/tree/master/CNN/HighPerformanceComputing/example/model)下载对应的[.param](https://github.com/Ewenwan/MVision/blob/master/CNN/HighPerformanceComputing/example/model/mobilenetv2_ssdlite_voc.param)文件和[.bin](https://github.com/Ewenwan/MVision/blob/master/CNN/HighPerformanceComputing/example/model/mobilenetv2_ssdlite_voc.bin)文件，然后修改[Detector.cc](https://github.com/Ewenwan/ORB_SLAM2_SSD_Semantic/blob/master/perfect/src/Detector.cc#L22)中的路径，替换为你的这两个文件的路径，然后保存、重新编译后，运行就没有这个报错了。
谢谢关注！ 本来应该是计划十月一更新的，但是因为导师催项目、催论文，可能我得等到寒假才能够补充完善了…… 而且可能我连寒假都没有了:joy:2333 等我找到空更新了后，再关闭这个issue吧
Yes! Same request. Even a closed-source SDK is also OK ... :sob: Please think about it, please! Few manufacturers have ACTUAL cross platform drivers (windows & Mac & Linux support...