The FOMO model runs perfectly fine on the Portenta, but when I try to light up the LED_BUILTIN, it gives the Failed to allocate TFLite arena error.
The example code gives the following solution:
** NOTE: If you run into TFLite arena allocation issue.
**
** This may be due to may dynamic memory fragmentation.
** Try defining "-DEI_CLASSIFIER_ALLOCATION_STATIC" in boards.local.txt (create
** if it doesn't exist) and copy this file to
** `<ARDUINO_CORE_INSTALL_PATH>/arduino/hardware/<mbed_core>/<core_version>/`.
**
** See
** (https://support.arduino.cc/hc/en-us/articles/360012076960-Where-are-the-installed-cores-located-)
** to find where Arduino installs cores on your machine.
**
I was successful in finding where it is located (even though the link does not work), but I am not sure how to edit the file, because the file expects a key=value format.
The Failed to allocate TFLite arena is a memory issue, you can try to use a smaller EI model (reduce your window size or your image size is usually a good start).
For the option Try defining "-DEI_CLASSIFIER_ALLOCATION_STATIC" in boards.local.txt. I’ll check with our Embedded team if they have a deeper explanation. I’ll add it to our documentation too.
Hi the URL looks to no longer be working. I’ve find the newer URL. You’d want to navigate to Board platforms and core and find the right path depending on your OS.
E.g. on Linux for the Portenta H7 using mbed core 2.8.0 and for the M7 you’d want to create boards.local.txt with the following path ~/.arduino15/packages/arduino/hardware/mbed_portenta/2.8.0/boards.local.txt.
The behavior you describe may be a memory fragmentation issue. Allocating the tensor arena may or may not solve the issue. It all depends on whether after allocating the tensor arena statically they’ll have enough memory for the rest of your application. Can you share the serial output when it fails and also the (led control) code snippet that causes the issue? And what’s the size of your model?
Starting inferencing in 2 seconds...
Taking photo...
Failed to allocate TFLite arena (error code 1)
ERR: Failed to run classifier (-6)
Following is the code for the led control:
if (bb.label == "bottle" && bb.value >= 0.90) {
Serial.println("Bottle found!!!! with " + String(bb.value*100) + "% accuracy");
digitalWrite(LED_BUILTIN, LOW); // further predictions failed after this line
}
I am not sure how to get the size of the model…can you tell me how to go about that?
Thank you. The model fits nicely in the Portenta. So I suspect your issue was due to memory fragmentation.
The combination of allocation/deallocation of the tensor arena (large area) with the alllocation/deallocation in your Serial.println("Bottle found!...") (small area) causses this. Did allocating the tensor arena statically solve your issue?
Please, could you assist with this case too?
I am using Arduino Nano 33 BLE Sense, and my model size is as shown below. I know that the RAM is 256kb for Nano BLE. Will this 254KB be the problem? Again, I have managed to locate the Boards.local.txt file, How can I redefine the “-DEI_CLASSIFIER_ALLOCATION_STATIC”
254K won’t work on the Arduino nano 33 BLE Sense with our firmware. 254K RAM is just for the model and doesn’t account for other parts of the firmware. The firmware uses more RAM more for the sensor(s), remote management and other processing, .etc…
Try reducing your model.
Another option, perhaps you don’t need all of the parts of firmware, is to write your own custom app. You can download the Arduino Library and work your way from there.
Thank you very much for your assistance. I was indeed running inferences beyond what the firmware could afford. After I followed those instructions, I now run inferences smoothly on my Arduino Nano 33 BLE sense.
