Tiny Machine L. Kit And OV7675 Camera Problem

Help
#1

Hello Edge impulse family
This will be my first project with Edge impulse and Arduino Tiny Machine learning kit.
Please forgive my excitement as I am new to Edge impulse and Tiny machine learning kit.
I created an edge impulse project that recognizes fruits. (project no:149197) and I came to the last step, deployment. In this step, I took a file for Arduino 33 BLE Sense and uploaded this file to the card on my windows machine with the node.js command by putting my card in boot mode.

Then when I give the command edge-impulse-run-impulse --debug on my screen;

Edge Impulse impulse runner v1.16.0
[SER] Connecting to COM4
[SER] Serial is connected, trying to read config...
[SER] Retrieved configuration
[SER] Device is running AT command version 1.7.0
Want to see a feed of the camera and live classification in your browser? Go to http://192.168.1.34:4915
[SER] Started inferencing, press CTRL+C to stop...

I get a message like, but when I go to the specified address, I can’t see any screenshots.
also when i use :edge-impulse-run-impulse command directly

Edge Impulse impulse runner v1.16.0
[SER] Connecting to COM4
[SER] Serial is connected, trying to read config...
[SER] Retrieved configuration
[SER] Device is running AT command version 1.7.0
[SER] To get a live feed of the camera and live classification in your browser, run with --debug
[SER] Started inferencing, press CTRL+C to stop...
LSE
Inferencing settings:
        Image resolution: 96x96
        Frame size: 9216
        no. of classes: 3
Starting inferencing in 2 seconds...
taking photo...
ERR: failed to allocate tensor arena
Failed to allocate TFLite arena (error code 1)
Failed to run impulse (-6)

I am getting an error of the form.
I can connect my card within the edge impulse website and upload images to edge impulse, but I cannot use the card alone.
What should I do to identify objects with the OV7675 camera? Where am I doing wrong?
Thank you in advance for your help.

#2

Hi @seloak,

The Arduino BLE Sense is limited in RAM (256 KB), your model already takes most of it. You could use static allocation to bypass this issue, see: GitHub - edgeimpulse/firmware-arduino-nano-33-ble-sense: Edge Impulse firmware for the Arduino Nano 33 BLE Sense development board
Otherwise choosing a smaller FOMO model may help too.

Aurelien

1 Like
Camera OV767X Error
#3

Thanks for help @aurel
What can I do for the model to use less memory?
-Like reducing the number of samples?
-Reducing the image size (like 36x36) ?
-Changing some parameters?
My current Object Detection param.
Number of training cycles
25
learning rate
0.015
Validation set size 20 %

I would be very grateful if you could explain what I need to change.
It would be great if there was an explanatory guide for Arduino BLE Sense.

Thank you for your help.

#4

Hello @seloak,

Are you using FOMO (Object Detection) or MobileNet transfer learning (Image classification).
Reducing your image size will definitely improve the on-device performances.

You can also reduce the color depth (choose grayscale), it will reduce the number of features by 3.
You can also choose a model using MobileNetV1 instead of MobileNetV2 and choose a smaller alpha if you selected 0.35.

Number of training cycles, learning rates and validation set size won’t have any impact.

Best,

Louis