ceres-solver | A Nonlinear Least Squares Minimizer

First check if you have access to the file:

After some more researching I figured out how to make this work, in case anyone stumbles upon the same problem. I ended up reading this issue description (link) and made me think I should try and use a different code generator. My cmake configuration was correct however apparently there is a bug with using XCode as the code generator (in here -G"$GENERATOR_NAME" ). After I set GENERATOR_NAME=Ninja I managed to compile a version of the library that is for Mac Catalyst.

I had the glog parameters in cmake set incorrectly.

Here is the correct setting:

• glog_DIR: Where is the build. There should be glog-config.cmake. Path: glog/out/build/x64-Debug
• GLOG_LIBRARIES: Where it is installed. Path to glogd.lib. Path: glog/out/install/x64-Debug/lib/glogd.lib
• GLOG_INCLUDE_DIR: Where it is installed. There should be folder named glog. Path: glog/out/install/x64-Debug/include

CubicInterpolator cannot handle non-uniformly distributed data. You will have to use something like a cubic spline yourself to do that.

The primary reason is that Ceres considers sparsity only at the level of parameter and residual blocks, not individual terms. For a truly dense problem, where every residual term depends upon every parameter, how the problem is represented has no significant impact on the runtime performance.

However, Ceres is designed to handle very large sparse problems where every residual term depends on only a few parameters, but there are a lot of parameters and residuals. It is absolutely critical to exploit the sparsity structure of these problems to be able to solve them efficiently - i.e. avoid a lot of pointless calculations where all the terms are zero. Whilst the linear algebra is mathematically identical, the approach by which the linear system is represented and solved is very different when dealing with sparse matrices.

A secondary reason, is that it makes modelling the problem easier. Loss functions are applied to the result of: r^T * r (where r is a residual block). If you have only a single residual block, the loss function is effectively scaling the total cost rather than down-weighting just the parts of the problem with large errors (outlier rejection). Ceres can also thread the evaluation of residual blocks, thus speeding up the evaluation of large problems.

The last part of your question is different. In the curve fitting example, there are two parameters and kNumObservations residuals, the problem is heavily overdetermined. However, if you instead had a single residual that internally calculated the errors for kNumObservations points, but only returned their sum, then the problem as solved has two parameters but only one residual and is underdetermined. As a separate note, returning their sum would also be different because you are not returning the squared sum, thus errors of different signs in different residuals could effectively cancel each other out - none of which would be observable by the solver.

It seems my build folder was corrupted. I just deleted the build folder & the install folder. When I re-run the install, everything worked perfectly.

“Killed” is what you usually see if something (external) sends SIGKILL to the compiler process. Either this is another user with permissions being bothered by this process or, more likely, the Linux OOM Killer evicting this process under memory pressure. I would suggest having a look at the memory consumption during compilation of “nist.cc”.

Yes, you can get an implicit conversion by defining a constructor that takes a single parameter (of the type you're converting from).

The issues lies in ConstPtr a(this); - this line creates a std::shared_ptr that takes ownership of the current object and deletes it after it goes out of scope.
Due to this the point pointer becomes a dangling pointer after the call float d2 = point->distance<4>(plane); and dereferencing it leads to undefined behavior

In order to fix the issue the Point class needs to inherit from std::enable_shared_from_this and the line ConstPtr a(this); needs to become ConstPtr a(shared_from_this());