leekeifon
leekeifon
@vrtmrz confirm that 0.23.6 version works as expected. One more question, can we start obsidian at the back(I mean without focusing on it) and make livesync to sync editions made...
@koide3 Thanks for the script! I have another question on how to set the "correct" `tag_size`. If the physical size of tag is 1mx1m, what's the recommended `tag_size`? 1024x1024? What...
> Usually you don't need to change `tag_size` (texture size) unless you see some blur on rendering results. 1024x1024 is sufficiently large for most situations. If you want to change...
> 自动微分进行优化 > 请教一个简单问题： > > 考虑标准BA问题中的一个误差项 e = ||u - 1/z_K_T*P||^2 其中，u表示像素坐标，K是相机内参，T是待优化的位姿，P是三维空间的点 > > 对位姿求导，我们手推增量jacobian的时候通常会转成李代数上来做，主要原因是当把T看成整体时，T中的旋转部分需要满足特殊的约束（只有良好的乘法运算，而没有加法运算）。但如果我们把这个误差项全部展开到矩阵上的每个元素，传统的微分方法是否依然可用？换句话说，是否可以不管旋转矩阵的约束，直接用自动微分方法计算误差e关于T中每个元素的导数？ > > 实际上，我看ceres自带的[ba优化例子](https://ceres-solver.googlesource.com/ceres-solver/+/master/examples/simple_bundle_adjuster.cc)中似乎直接用自动微分进行优化，没有显式的写出jacobian。此外，另一个问题是，例子中也没有显式的定义T的更新规则，那么ceres是如何知道用T=\Delta T *T而不是T=\Delta T + T来进行更新？ 你的疑问解决了吗？我只理解了这个代码逻辑成立的一部分。 1. 增量是可以直接使用向量加法更新状态量的，因为旋转用的是angle-axis representation 2. 但是，autodiff最后是用numeric...
Yeah, though this seems like a bug, but I don't really test it out. Any idea on how to verify the fix is correct?
@SteveMacenski OK, I guess I know how to prove it a bug. Here's the failure mode: Set some extreme param ``` karto::Mapper* pMapper = new karto::Mapper(); pMapper->Reset(); pMapper->setParamCoarseSearchAngleOffset(0.255); pMapper->setParamCoarseAngleResolution(0.017); pMapper->setParamFineSearchAngleOffset(0.051);...
I get hang too, not sure why
Already verified, once I comment out the `add_subdirectory(tools)` line first, and compile the code, the messages will generate successfully.
Also, can't build objects with g2o successfully. I try two ways to build the entire project. 1. build at the {project_src}/build, and trigger an error 2. I run `cmake ..`...