Wondering if the Caffe tutorial can be applied for iMXRT1060 OpenMV Cam?
Hi, that’s very old. Just use Edge Impulse. The ability to deploy a tensorflow lite network and have it run is mature.
Thanks for the reply… We are looking for some segmentation solutions and edge impulse does not offer it. Moreover, the tensorflow library that openmv forked was too old and since there were lot of bugs fixed. Any ideas would be appreciated. Thanks!
This lib is too old? openmv/tensorflow-lib: TensorFlow package for Cortex-M0-Plus, Cortex-M4, Cortex-M7, and Cortex-M55 CPUs. (github.com)
For the RT1062 we will be enabling all ops. Just waiting on this PR to get merged once Ibrahim has time: modules/py_tf: Refactor regression() code to handle arrays correctly. by kwagyeman · Pull Request #2133 · openmv/openmv (github.com)
You should be able to run any TensorFlow network then. I can confirm being able to run YOLOV3/V5 with the latest updates to be merged into the firmware.
Wow… this is awesome… are you talking about the segmentation models or object detection one?
Uh… I’m just saying that if you want to run a segmentation model it shouldn’t have a problem. That’s what Edge Impulses FOMO network is.
Things may have not been working because we disabled a bunch of ops on our cameras to make TensorFlow fit. However, with the RT1062 we will enable all ops on the board once the above PR is merged.