Portenta H7 and Vision Shield not working with Machine Learning Examples

Hi there!

Just got myself a Portenta H7 and the Portenta Vision Shield and am working my way through the examples but realised the examples under Machine Learning - Tensorflow doesn’t work with this set up as the models aren’t preloaded onto the board so it can’t access them? Wondering how can I get access to those models so that I can run them - mainly looking at the YOLO models and Blazeface as well as Face landmarks

Thanks!

Hi, you need to install the ROMFS. Just go to Tools→ROMFS→Reset ROMFS.

This will load blazeface and yolo_lc onboard which can run at about 2 FPS. Not exactly real time, but, usable, and the models are much better than anything we had before.

As for facial landmarks, that really needs an NPU, you can try to run it on the Portenta but you will get sub 1 FPS. To install it…

  1. Tools→ROMFS→Edit ROMFS
  2. Open the Model Zoo from the dialog
  3. Uncheck the filter models by board type
  4. Go into STMicroelectronics Folder→Pose Estimation→head_landmarks
  5. Then select the model and copy it over.

Again, only YOLO_LC and BlazeFace will run at a really usable FPS. But, you are free to try the other models. When the N6 and AE3 hit the market, they can run everything at real-time framerates.

Also, the models like color images… not sure how the grayscale image will affect them from the Himax camera. But, be aware of this.

Hey!

Thanks for answering the question! I managed to find it (Edit Romfs on OpenMVCam right? )but whenever I click on a model the IDE crashes so I haven’t been able to try it out. Any tips around that? I’m on OpenMV IDE 4.8.1 and MacOS Tahoe 26.2

Thanks for the quick reply btw!

Hi, can you send the stack trace?

Also, I’ll be updating my Mac Mini M4 test machine to Tahoe 26.2. It appears Apple changed quite a bit about how the OS works under the hood in the latest update to the 26 naming scheme. I had a customer over to my house on Tuesday to debug things, and we realized Apple had wholesale changed the stable command-line tools in the OS, etc. I’m going to run through all the features of Mac to make this stop happening.