eIQ Machine Learning SoftwareのトピックRe: What about object detection on i.MXRT?

What about object detection on i.MXRT?

MarcinChelminsk — Mon, 24 Feb 2020 15:24:01 GMT

Hi NXP Team!

I am aware that I can import example from SDK called "tensorflow_lite_label_image" (which is example for image classification). I followed the lab called "eIQ Transfer Learning Lab" which is based on this SDK example (https://community.nxp.com/docs/DOC-343827 )

However, the question is: what about object detection on i.MXRT? Do you provide any examples, guides?

Any hints more than welcome! Thanks in advance!

Re: What about object detection on i.MXRT?

david_piskula — Thu, 27 Feb 2020 09:59:05 GMT

Hello Marcin, we do not currently have an object detection example for the i.MX RT devices.

Re: What about object detection on i.MXRT?

jkjung13 — Mon, 31 May 2021 06:33:58 GMT

I'd like to raise the same question again, since it's been over 1 year since the last post. Does NXP have an object detection example for the i.MX RT devices now? Thanks.

Re: What about object detection on i.MXRT?

Ramson — Thu, 22 Jul 2021 11:10:18 GMT

Hi, we are also working on a similar object detection example for tensorflow lite in which we are facing an issue, which i have posted in the forum (https://community.nxp.com/t5/eIQ-Machine-Learning-Software/How-to-add-custom-operators-in-tensorflow-Lite/m-p/1311948). It would be a great help if you can find any solution for this, so that we all can benefit. Thanks

Re: What about object detection on i.MXRT?

MarcinChelminsk — Thu, 05 Aug 2021 12:30:12 GMT

Hello,

as eIQ toolkit is avialable now, and it supports working with Classification and Detection models

could you provide any guidance on the workflow how to use Detection models and then deploy on embedded side?

Re: What about object detection on i.MXRT?

david_piskula — Wed, 11 Aug 2021 09:34:26 GMT

Hello Marcin,

with eIQ Portal there aren't any huge differences in the flow between classification and object detection. The main difference from a user perspective is labeling the various objects (instead of whole images) in the dataset curator. When your dataset is ready, you just need to choose an object detection model and afterwards the workflow is the same.

If you have any specific questions, please create a new standalone thread so that we don't keep reviving this old one. Let's discuss everything in a fresh one with a more specific subject so that it's easier to find for others who look for the same information.

Regards,

David