sensor.set_framesize(sensor.SVGA) is the max setting for openmv H7plus.
if framesize is bigger than svga, the system will report tha buffer stack memory.
i want to know if openmv runs with set_framesize(FHD) and speed is 1 FPS， what the size of openmv RAM should have and how to set the stack?
yes, 1 FPS.
and many example of openmv set the framesize is very small, i think it is very bad for practical application。
The algorithm just requires a ton of RAM to work.
Remember that the camera is a microcontroller, typically you want a higher end application processor to do these things. Even then they won’t get that much improvement in the FPS.
Our next gen systems will be able to do much better on this, but, still a linear speed up on non-ML algorithms. ML algorithms will get an exponential speed up.
there are many linux board in taobao.com.
for example:
you don’t worry about that. openmv have lot of advantages.
openmv vs canmv 线段查找：canmv报堆栈错误。内存2G还不够吗？-CSDN博客
i hope next openmv will break limit of buffer stack for find_line_segments. 
We doubled the RAM on the next version, doubled the RAM speed, doubled the CPU speed, and then the CPU support vector instructions, which can quadruple algorithm speed after we implement optimizations to support for them.
So, initially, 4x faster, and then once the system hits maturity like 16x faster.
4*32MB=128MB，why not 1GB？
the progress is still small. i think the step coule be a bigger .
胆子大一点，步子快一点。
i don’t want to see out of fast frame buffer satck… when i setframesize=FHD for find_line_segments.
can i get that? and when?
Hi, find_line_segments() is just the LSD algorithm that you can run with OpenCV. Buy a linux SBC if you want that via high res. We explicitly build microcontroller cameras. The goal for them is low-power consumption and agility in being able to turn off/on quickly.
For the next gen product launch, you’ll be able to run desktop grade ML models onboard which can solve the problem you are trying to use find_line_segments for.
i have no idea about the ML. wish you have example of ML which i can learn step by step just i learn the openmv 4H7plus.
best wish for you!
Yes, we’re doing some partnerships with serious companies who can make it super easy to train ML models. So, it’s going to be good.
that is a good news! hope the bigger RAM and flash come early.
