Thsi is a very straightforward question, and hopefully there will not be roundabout / indirect answers. Surely, everyone that has used this software is aware that its GUI is severely broken. That is, completely dysfunctional in some very critical ways. We are coming from an open-source pipeline that is well streamlined but mostly headless. We wanted to use the MATLAB GUI as a testing ground for basic functionality. Unfortunately, it would appear the GUI doesn't work in either Linux or Windows (we have tested multiple computers and architectures), and the issues don't stop at a lack of GUI functionality but, apparently, in engine agility / performance as well.
Case in point -- the toolbox of interest -- Lidar Camera Calibration -- allows the specification of image and point cloud folders, but processing just a single pair takes roughly 5 minutes on a 32-core CPU with 8Gb RTX 2070 GPU. We use this same workstation to run very elaborate machine learning algorithms with big data analysis. All of the GPU intensive workflows that manipulate multi-million point clouds, ICP matching, etc. are able to process hundreds of clouds in minutes --- meanwhile, MATLAB is laboring over single pairs for up to 1 hour (for 50 pairs). Even just 4 pairs will consume over 10 minutes. This kind of performance is absolutely unacceptable.
But that is not what I am opening this query about. Even once the pairs have been analyzed, MATLAB apparently isn't sophisticated enough to automatically identify a checkerboard plane, for example, in the point cloud that corresponds to a 2D image --- something that PCL, for example, can do with absolutely no issue using RANSAC. Moreoever, in the graphical interface, it is not possible to edit the ROI. There is an "Edit ROI" button, but the software is so unconscionably slow that just changing from rotation to panning or zoom will invoke a 30-60 second delay, and there is absolutely no possibility for doing anything whatsoever with the bounding box around the points. Moreover, (and perhaps more surprising than anything else) there is no detailed documentation for this interface.
I understand many of the "veterans" here are using the command-line interface exclusively, but I see no need to take on a new language when the platform itself grossly underperforms literally every other GPU-based open-source project out there. I wanted to give MATLAB a chance to show its capability for computer vision, but this first impression, unfortunately, will be quite difficult to reverse. I originally thought the lack of performance and efficiency in the engine and GUI were Linux based, so we moved off the machine learning laptops and onto the more robust desktops. The same behavior is seen regardless of system or OS -- so it's clearly a design flaw.
We were interested in purcashing 100+ licenses for our research branch, but it's clear, both from reading the complaints here in the forums, and observing the overall benchmarks conveyed in other threads -- that it would have been a costly mistake. I wanted to leave this post here, just on the odd chance that someone -- somewhere -- has run into these issues and is aware of of a workaround.
I doubt many people are using the Computer Vision functionalty of this suite, or else these problems would likely be more visible and addressed in a timely fashion. Perhaps it is a limitation of 2021a. I cannot say, because this is our first (and last) excursion with MATLAB. Hopefully, the developers monitor these threads and will take some honest and well-intentioned criticism for the product, in general.
My suggestion would be to open-source this software, given that it has not yet achieved commercial-level stability. It's very far from being a product worthy of even a meager price-tag, with these obvious and unacceptable flaws, and I can't imagine Mathwork's profit margin can be even remotely robust -- unless there are more people out there than I would care to imagine that are OK waiting hours for their algorithms to run rather than seconds with open-source. That leads to another unrelated question, "why pay for this when you can do better for free", but this is clearly not the right place to post such a question. I am open to the possibility that there may be ways to optimize the engine (but doubtful, since such optimizations would be expected to be automatically engaged as part of the base installation).