Can I run Opencv in tflm_cifar10 with MIMXRT1060-EVKB?

[nnxxpp]

My question is in the title. I want to use some built-in method in Opencv but I am not sure, whether I can run Opencv on MIMXRT1060-EVKB.

If so, please help me to config it. Where I need to put source code and config to build and flash to the board.

Thank you.

[nnxxpp]

@Sam_Gao @Dennisleon 

Thank you. I understood. I am a newbie with embedded.

My case is that I build TF model on PC with image resizing by OpenCV. When deploy on MIMXRT1060-EVKB, I use examples from SDK (tflm_cifar10) and it has simple image resizing ==> accuracy of model is not good as on PC. It is reason that I want to use OpenCV on the board. Actually I make use of Makefile of tflm_cifar10, I don't know how to integrate Opencv in this project.

Could you give me some advice to get better accuracy on the board? Thank you.

[nnxxpp · Accepted solution]

@Sam_Gao @Dennisleon 

Thank you for useful information. 

OpenCV is the powerful tool. In my case, I want to use some resizing methods in OpenCV. 

Due to memory constraits and more complex compilation procedure, I want to get another solution. I found that this repo https://github.com/nothings/stb/blob/master/stb_image_resize2.h . It is easy for me. Simply, I put .h file to the source code and run it without changing any configs.

[Sam_Gao]

Hi @nnxxpp 

Thanks for your question, many OpenCV examples running on PC or other MPU/SoC/AP platform mostly, there is some referece application note about how to build an openCV library with GCC/ARMClang and run it on our RT-series MCU Platform (e.g. RT 1170 EVKB board), I strongly recommand to get more from this AN which base on our XPresso IDE: https://www.nxp.com/docs/en/application-note/AN13753.pdf, also example to deploy on the 1170 EVK https://www.nxp.com/docs/en/application-note/AN13725.pdf

Have a nice day!

Sam

[Dennisleon]

Hello @nnxxpp ,

The i.MX RT1060 Evaluation Kit (MIMXRT1060-EVKB) is a powerful platform for evaluation and prototyping using the MCUXpresso suite of software and tools. It features the i.MX RT1062DVL6B processor, which is a Cortex-M7-based crossover MCU. Here are some details about the board:  

Processor: The MIMXRT1060-EVKB is equipped with the i.MX RT1062DVL6B processor.

Memory: It has 256 Mb SDRAM memory, 512 Mb Hyper Flash, and 64 Mb QSPI Flash.

Interfaces: The board includes various interfaces such as USB host and OTG connectors, Ethernet (10/100M) connector, CAN transceiver, and an Arduino® interface.

Display and Audio: It features a parallel LCD connector, camera connector, audio codec, 4-pole audio headphone jack, external speaker connection, and a microphone.

Software Support: You can develop applications using the MCUXpresso SDK with optional FreeRTOS support and the MCUXpresso IDE. Additionally, it is supported by Zephyr OS.

To get started with your MIMXRT1060-EVKB, follow these steps:

Configure Boot Mode:
The board has four boot modes, with one reserved for NXP use. To boot from the QSPI flash, ensure that SW7 is set to 0010.

Attach the USB Cable:
Connect the MIMXRT1060-EVKB board to your PC using the included micro-USB cable (connector J1).

Run the Out of the Box Demo:
Your MIMXRT1060-EVK comes preloaded with a “LED blinky” demo that uses the onboard LED. When the board is powered, the Green LED should flash on and off.

Regarding OpenCV, the MIMXRT1060-EVKB is primarily designed for embedded applications, and running full-fledged OpenCV might be challenging due to resource constraints. However, you can explore using OpenCV Lite, which is a lightweight version optimized for embedded platforms. Here are some steps to consider:

Cross-Compilation:
Set up a cross-compilation environment for the MIMXRT1060-EVKB. You’ll need a toolchain compatible with the Cortex-M7 architecture.

Build OpenCV Lite:
Obtain the OpenCV Lite source code and configure it for your target platform. Compile it using the cross-compiler.

Integration:
Integrate the compiled OpenCV Lite library into your application project.

Test and Optimize:
Develop a simple OpenCV application (e.g., image filtering) and test it on the board.

Optimize your code for memory and performance constraints.

Keep in mind that you’ll need to manage memory carefully, as the MIMXRT1060-EVKB has limited resources compared to desktop systems. Additionally, consider using hardware acceleration (if available) to improve performance.

I hope the information may helps you.