Why are there differences between my scale and your app?

We are constantly improving our AI to handle different field conditions. Stay tuned for updates coming soon to better handle shrimp on their side and the ability to get accurate results without water in the tray.

You may notice differences between MAX’s average body weight of a sample and your scale. This is normally caused by issues with images submitted to xpertSea MAX during a monitoring, and can easily be corrected in future samples!

How does xpertSea get an average body weight of the shrimp in a sample?

Let’s first outline how xpertSea MAX calculates the average body weight of a sample. 

  1. Identify shrimp: MAX uses AI to identify every shrimp in xpertSea’s trays. This AI relies on finding the white background of the tray behind each shrimp to mark and outline the shrimp.

  2. Exclude bad shrimp: Each tray sampled may have 1-2 shrimp that cannot be detected or weighted. To keep sample results accurate, MAX excludes these shrimp from calculations.

  3. Infer the weights: MAX then associates each shrimp in the image to shrimp in xpertSea’s database of over 4.5 million images to calculate the weight of the new shrimp.

When is it likely that I will see differences between my scale and xpertSea MAX?

Below are the most common reasons why there may be differences. Check out our <field photo taking guide> on how to take the best photos for accuracy! 

  1. Comparing results using different animals. Comparing different shrimp using the scale and MAX will lead to different results. If comparing weights, ensure you are using the same exact shrimp in each sample. Check out how you can use MAX to identify and prevent sample error to learn more about which animals to sample.

  2. Shrimp are overlapping in the tray. MAX’s model can’t get an accurate weight if there is no separation of where one shrimp ends and another begins. In the example below, you can see how a tray with shrimp that are clustered and overlapping leads to inaccurate identification results:

3. When shrimp are curled up on their sides, without any water. MAX’s model was trained with live shrimp that are not curled up on their sides. If shrimp are without water or curled up, MAX’s model may overestimate the weight. 

How to get the best results in your sample

Here are three quick tips on how to take a better image:

  1. Make sure you can see the white background of the tray around every shrimp in your sample. Review images regularly on your phone or on the portal to see if your workers are following these instructions.

  2. Use fewer shrimp per tray, especially at larger sizes. At smaller sizes (<15g), you may only need 2 or 3 pictures to reach the minimum recommended sample of 150 animals. At larger sizes (>15g), you may need 6-8 images per sample.

  3. Use enough water so that the most shrimp are upright. If you are not using water, your results are likely to be over-estimating the weights.

How can MAX help?

xpertSea is constantly improving its weight algorithm, but will also warn you when there is potential for sample error using its prediction curve. If you are unsure of the weight, it is more likely that the weight prediction for that day is more representative of the pond’s weight than the sample’s weight. Check out how you can use MAX to identify and prevent sample error to learn more about using predictions in your planning.

Finally, if you have examples of ponds where results are highly variable, check the images and reach out to your xpertSea representative to discover changes you can make in your process for better results!

Previous
Previous

What are the device requirements?