Iterative Training

Model training is an iterative process. As you begin testing out real world examples on your model, you can identify weaknesses and use test results to pinpoint areas for improvement.

The first thing to adjust to improve your model output is the Confidence Threshold. If your model didn’t find an object you wanted it to, see if it picks it up with a lower confidence threshold. If your model is predicting too many of your object and you’re getting a lot of false positives, see if you can increase the confidence threshold to get better results.

If you’re not able to balance the confidence for these two scenarios, begin to analyze the image you’re working with. First, use the button to send this image to your overall dataset so that you can train your model on this specific image the next time around. Then, go out and find similar images to add into your dataset. Think about lighting conditions, camera angle, object size, background noise, etc. Think through what might be confusing for the model and bring in more similar examples.

Repeat this process to enhance accuracy and reliability until you are satisfied with your model results. Congrats! You’ve done it.