Is there a way to adjust the target device parameters to take advantage of the extra memory (practically unlimited, given processing time constraints) available on OpenMV’s H7 Plus, where the 32MB framebuffer can be used to store a TF model. While a model that large couldn’t be processed in anything like real time (and some space is needed as a framebuffer), it does effectively eliminate the space constraints.
The STM32H743II also has a faster clock than the predefined Cortex M7, so it would be nice to adjust that properly too, though scaling the processing time target is easy enough.