Skip to main content

Create the plastics dataset

Create and name the dataset​

  1. Open Datasets.
  2. Select New dataset.
  3. Rename the new dataset to Nylon and Belt.

Assign the image​

  1. Open the dataset's Images tab.
  2. Under Available data, select the uploaded image.
  3. Set Data assignment method to Automatic.
  4. Select Add to bucket > Training & Validation.
  5. Confirm that the Plastics labels set is assigned. Only a label set with Finalized status can be used.

Automatic assignment splits eligible labeled data between training and validation according to the displayed Data split control.

This one-image Automatic split teaches the product workflow; it is not an independent generalization benchmark. With a Pixel split, correlated pixels from the same labels and image can enter both buckets. For a scientific evaluation, use separate images or acquisitions and Manual assignment.

Configure the draft​

On Overview:

  1. Set Classes to Belt and Nylon.
  2. Set Data level processed to Reflectance.
  3. Select Wavelengths and choose the center-wavelength set for the uploaded image.
  4. Review the sampling and split settings.

Finalize​

Select Finalize. If the action is disabled or validation fails, check the classes, wavelengths, processing level, data assignment, and finalized label-set assignment.

After finalization, review the dataset metrics and average signatures. Distinct-looking signatures can be a useful quality check, but they do not by themselves prove that the labels or future model are correct.

Next: Train the model.