Hi Everyone,
I am trying to use the LSTM model that is trained with Keras and converted into the TFLite in OpenMV. Whenever I loaded the model into the OpenMV, it blinks white repeatedly then reboots (I assume that because of the unsupported layers). Is there any way that I can run the model in OpenMV. The architecture is below.
Yeah, the tensorflow code base just segfaults on model ops that aren’t supported. I have a task to merge the updated library here: tensorflow-lib/libtf.cc at master · openmv/tensorflow-lib (github.com)
Please verify all the ops you need are in that list. If they aren’t the system will just reboot like it’s doing on looking for a missing op. EDIT: Looks like all the ops you need are supported by the new library. So, I just need to put it into the firmware.
(The segfault is from Google code. Yes. I know.)
Anyway, since the jpeg hardware decoder PR is now merged we have firmware space for this updated tensorflow library. I can work on this over the weekend to send a PR to merge. I can also generate a new firmware for the camera for you until the PR is merged once I’m done with the code changes.
First of all thank you for your answer and effort. If PR will take a long time, it would be better for me to send the new generated firmware if possible.
Hi, yes, I will be working on this over the weekend or into next week. It will get done very soon though. I’ll send the new firmware once I have this complete.
