Dear NXP,
I'm masters graduate student and doing my academic project using i.MX RT1060 Evaluation Kit.
The objective of the project: is to implement "academic attendance system using face recognition". In order to reach my objective, I used eiq examples from SDK_2_6_0_EVK-MIMXRT1060 --> evkmimxrt1060_tensorflow_lite_label_image as a base software. At present I've integrated retrained mobilenet_v1_0.25_128.
Problems I'm facing:
1) I'm struggling to get accuracy with my retrained model when I use real time data captured by camera which came as part of the kit (please note, I've adjusted the focus to get maximum clarity of the image). I get almost 100% accuracy when I give static images for recognition while software is running on the board. I retrained the model by using datasets captured by demo project "evkmimxrt1060_camera_capture" where the images were captured by camera with EVK in which I'm using for test. other data sets were captured using computer webcam. In both models, there is no improvement in accuracy.
2) Current demo projects are reading the face/object from fixed location. But I want to detect face dynamically and crop the data for face and pass it to tensor library for face recognition.
Support needed:
1) Are there any other demo projects with models (along with document to retrain models) for face recognition ? Could you share base project as I don't have enough time for final submission.
2) Is there any document which guides me to create open CV mini library or lite open CV library cross compiled which suits to run on I.MXRT1060 also document which lists interfaces to use?
I also ordered SLN-VIZN-IOT to explore but I could not develop the intended application which I need to submit, As the demo application is in binary format. which have core face recognition algorithm.
Could you please help with solutions for my project.
Thank you!
Priyanka
