Skip to main content

Inference Results

Overview​

Once your model meets your performance requirements, you can use it to generate inference results for any image that partially matches the center wavelengths of the model. Simply upload the image to Clarity Drive, open it in Spectral Explorer, right click it, and select "Generate results" to run your image on the model.

For best performance, the new image should be captured using similar center wavelengths as the one used to train the model. While the data can be resampled to match the model's center wavelengths, this may introduce noise and reduce the quality of the results.

Using the Finished Model​

Running Inference​

Step 1: Prepare Your Image​

If you haven't already, upload the new image to Clarity. Make sure the processing level matches the processing level of the model.

Step 2: Generate Results​

  • Open the image in Spectral Explorer
  • Right-click on the image
  • Select "Generate results"
  • Choose your trained model from the list

Step 3: View Results​

  • Results will appear in the Results tab
  • Visualize results directly in Spectral Explorer
  • Export results for further analysis

Viewing Inference Results​

You can see inference results either in the results tab of the models page, or in Spectral Explorer.

In Models Page​

  • Navigate to the Models section
  • Select your trained model
  • View results in the Results tab
  • Compare performance across different images

In Spectral Explorer​

  • Results overlay directly on your image
  • Adjust visualization settings
  • Compare with ground truth or other results
  • Export results for external analysis

Result Types​

Classification Results​

  • Color-coded pixel classifications
  • Confidence scores for each class
  • Class probability distributions

Unmixing Results​

  • Abundance maps for each endmember
  • Confidence thresholds
  • Material composition percentages

Target Detection Results​

  • Target probability maps
  • Detection confidence scores
  • False positive filtering options

Regression Results​

  • Continuous value predictions
  • Uncertainty estimates
  • Statistical summaries

Best Practices​

  • Validate results against known ground truth when possible
  • Check sensor compatibility before running inference
  • Monitor confidence scores to identify uncertain predictions
  • Use appropriate thresholds for your specific application
  • Document inference parameters for reproducibility