Change Number of Labels

Help
#1

I create a Classification project, creating a dataset of images taken with a Nicla Vision board. But pretty much screwed that up, so I deleted the dataset and created a new one with fewer labels. When I went to Impulse Design I changed the Output Labels, but the Model Output remembered I had more labels and wouldn’t let me save the new model. I exported the new dataset, deleted it from the project, then re-imported it. It still remembered the old number of labels. Can I fix this or do I have to create a new project?

#2

Hmmm. Went away to sulk. Came back later and added a couple images w/ new labels, went Impulse Design (Create Impulse) and it had the right info. Didn’t even show the Model Output box. Happier now.

#3

This problem reappeared on another(?) project. I have a project (Martin/Snoring), basically cloned form a Hackster project that worked in detecting snoring noises from non-snoring. I versioned it as V2. I may have played with it afterwards to use spoken numbers (and hence my original problem). Later went back and it had the snoring dataset (2 classes: snoring and non-snoring) but going to Impulse Design put me on “Step 2: Process saved_model.zip:” and 10 output shapes (10 digits). Can’t figure out how to tell it I only have 2 output shapes. The V2 version still works (almost) correctly (maybe subject of another post). Any help?

#4

Hi @mjenkel

This can happen when starting out,

Here are steps you can take to resolve the issue and ensure your project aligns with the updated number of labels:

  1. Verify Project Configuration:
  • Ensure that your project’s dataset accurately reflects the current labels you intend to use. You can do this by visiting the “Data Acquisition” section of your Edge Impulse project and reviewing the labels associated with your data samples.
  1. Recreate the Impulse:
  • Navigate to the “Create Impulse” section and adjust the output features to match your current labels. Removing and then re-adding the “Learning Block” with the correct number of output classes may force the system to refresh and acknowledge the updated label configuration.
  1. Retrain the Model:
  • After adjusting your impulse configuration, retrain your model. This process should take into account the updated number of labels and adjust the model’s output layer accordingly.

Hope this helps rectify your issue,

Best

Eoin