Object Tracking Post-processing

[rem123]

Question/Issue:
How can I deploy the object tracking post processing on the Grove Vision AI V2?

Project ID:
788304

Context/Use case:
I am training a vehicle and motorcycle detection model. I though when I built the binary for the Seeed Grove Vision AI Module V2 (Himax WiseEye2), the post-processing comes with it in deployment.

Steps Taken:

1. Saved my parameters for object tracking in post processing
2. Built the binary for the board
3. Deployed the model

Expected Outcome:
Tracking of objects as prediction results

Actual Outcome:
The typical object detection results

…Predictions (DSP: 5.15399 ms., Classification: 4.96899 ms., Anomaly: 0ms.):
Predictions (DSP: 5.15199 ms., Classification: 4.96899 ms., Anomaly: 0ms.):
motorcycle (0.65624) [ x: 112, y: 48, width: 8, height: 8 ]
motorcycle (0.60546) [ x: 128, y: 80, width: 8, height: 8 ]…

Reproducibility:

• [X ] Always
• [ ] Sometimes
• [ ] Rarely

[rem123]

Hi, Im still stuck with this one. Any help or answers? Thanks.

[Eoin]

@rem123

788304

Hi @rem123

Did you update the firmware of your Grove Vision AI to the latest? Can you also check your project export for the following?

In file:
model-parameters/model_variables.h
Check for:

const ei_object_tracking_config_t ei_posprocessing_config_9 = { 1, /* implementation_version */ 5, /* keep_grace */ 5, /* max_observations */ 0.5000f, /* iou_threshold */ true /* use_iou */ };

Object tracking - Edge Impulse Documentation

[Eoin]

Hi @rem123

I’m after checking your project, and you are using a Custom Block, this must be the source of your issue. However I’m not 100% sure, are you already in touch with our solutions and sales team?

They would be best placed to review and advise on your custom block work, I wrote the doc for Object Tracking so can help with any questions regarding calling it, but not with implementation on a custom deployment block unfortunately.

If not then reach out:
https://edgeimpulse.com/contact-sales

If you’re using this for a product or platform etc you would be best to have a conversation with them earlier rather than later too, so you get the best support in your effort. As this is what they work on with our customers.

Hope this helps!

Best

Eoin

[rem123]

Hi, thank you for the response. We initially exported the project with the supported binary deployment for the board. Upon checking the version of the firmware in github, there is a new version released two days ago. We will update the deployed firmware and I’ll make another reply.

This is how we initially exported the project. The custom block you see as our current selected deployment was just us looking for possible solutions for the post processing to appear. Because another model deployment method for this board is through the SenseCraft website by Seeed Studio where we can upload the .tflite file to it and see results. And the custom block deployment have that model file. However the post processing results are still not able to show.

For the model_variables.h, the supported binary deployment (that one above) does not have that folder and files. These are the files from the build.

Since I also have the custom block deployment build, I checked for:

const ei_object_tracking_config_t ei_posprocessing_config_9 = { 1, /* implementation_version */ 5, /* keep_grace */ 5, /* max_observations */ 0.5000f, /* iou_threshold */ true /* use_iou */ };

And yes, this exist in the files.

However, our plan is to use the supported binary deployment for the board. We hope there is a way for the post processing block to work with it.

Again, I’ll give another update after we deployed the latest firmware. As for the solutions and sales team, we are not in contact with them yet.

Thank you.

[rem123]

Hi, we updated the firmware and tried to redeploy the model. This is the results we are getting.

[rem123]

Update:
When pasting the vela model file from the supported binary build for the board to the pre compiled firmware from Seeed Grove Vision AI Module V2 (WiseEye2) - Edge Impulse Documentation, then following the readme instructions, these are the results:

The warning is gone, but it seems like the model is not working because it cannot recognize anything. This is also the case when edge-impulse-run-impulse --debug is ran. It only shows the live camera feed but does not recognize objects.

When uploading the model solely from the supported binary build, this is the output:

The result is just like from the previous reply. By the way, this is the latest build from the project’s deployment. The binary was built yesterday, Nov 1, 2025.

Builds from an earlier date (before firmware updates) show results. For example, this is the output when deploying a build of the project from October 22, 2025:

It detects objects, however no object tracking. Just like the initial problem of this thread.