i.mx8mp NPU accelerated segmentation demo

[fjpmbb_abc]

We used Nvidia Jetson nano segmentation demo using SegNet and we are interested in the case using the Cityscapes dataset.

The model file should be included in the Jetson package.

https://github.com/dusty-nv/jetson-inference/blob/master/docs/segnet-console-2.md

Now we want to use NXP i.mx8mp for processor replacement,
1. Is is a similar real-time, NPU accelerated demo on the i.MX board, with camera stream as the input?
2. Is it feasible i.mx8mp runing the above segmentation demo code?

Thanks.

@xiaodong_zhang 

[hiep_nguyen]

Hi,

Just share a bit my experience, the i.MX 8MP NPU can run a segmentation model over 15 FPS with 720p image resolution.

Hope it motivates you to keep experimenting.

[at_akihito]

@hiep_nguyen 

Hi.

I'm also trying to create a segmentation application that will run on a board with i.MX 8M Plus.
I'm generating segmentation TFLite models based on the existing TF SavedModel with conversion to TFLite, but they all only run at about 0.3 FPS.
Could you please tell me how to create a model that achieves 15 FPS at 720p?

Thanks.

[xiaodong_zhang]

@DevinJiao 

  Please have a look at the issue from @at_akihito .

[DevinJiao]

Sure. @at_akihito , can you share the model you used? Did you quantize it before you run it on NPU?

[at_akihito]

@DevinJiao 

I apologize for the delay in replying.

I have completed the model at a somewhat satisfactory speed, but another problem has arisen, so I will ask the question in another thread.

Thanks.

[xiaodong_zhang]

@fjpmbb_abc 

1. Is is a similar real-time, NPU accelerated demo on the i.MX board, with camera stream as the input?

Please refer to https://pyeiq.dev/ which supports camera.
2. Is it feasible i.mx8mp runing the above segmentation demo code?

The examples in eIQ don't include SegNet. But I found someone implemented it by Tensorflow. Customer can refer the below examples and convert TF model to TFlite which has deployed on i.MX8M Plus.

https://github.com/toimcio/SegNet-tensorflow

https://github.com/Keyird/TensorFlow2-Segmentation

[fjpmbb_abc]

@xiaodong_zhang 

   Thanks for your reply. 

   With your suggestion, i installed the pyeiq in the imx8mp board. But i can't run the demos and apps which pyeiq mentioned "pyeiq --run xxxx"

 The board said there is no valid server to download the demos and apps.  Whether the demos and apps need special network connection for download.(such as can visit the google.com)

[xiaodong_zhang]

@fjpmbb_abc 

Please see:

https://pyeiq.dev/3_faq/3_0_faq.html

[fjpmbb_abc]

@xiaodong_zhang 

   After reading the eiq FAQ you mentioned, i don't know what is the problem in my side.

   my i.mx8mp can visit the internet well as attachment show. But can't run the " pyeiq --run xxxx" well, encounter the "no servers were avalible to download the data".

[xiaodong_zhang]

Hi @fjpmbb_abc 

    Check with owner of PyeIQ, server has some issue. He is working on fix it.

[Bio_TICFSL]

Good

[fjpmbb_abc]

@xiaodong_zhang 

@Bio_TICFSL 

It seems like the pyeiq server still can't access.

May i know when this issue will be solved? Thanks.