Even though the computer vision model has been trained to work in field conditions, some situations can cause challenges. Below are some examples of issues with images that can lead to inaccurate results.

Clustering

When shrimp are placed in the tray with water, they often move toward the corners of the tray. When this happens, it can be impossible for the vision model to identify each shrimp. As can be seen in the two images, the shrimp are very clustered together, and the AI has a difficult time (as would any person) visually identifying each shrimp. This can lead to inaccurate results. Make sure you can see the white tray behind each shrimp and that no shrimp are overlapping.

To prevent this, stir the shrimp in the tray right before taking the picture. If this does not work, it may mean you have too many shrimp in the tray and need to put less in.

Sun/light reflection and bad lighting

In very sunny conditions, the water can reflect the sun and cause issues. As well, bad lighting such as partial shadow over the tray or a really dark image can cause issues in getting accurate results.

To prevent these issues, it is recommended to monitor early in the morning before the sun is strong. Alternatively, you can take the photo in a shaded area or use an umbrella to create the necessary shade.

Note: we are actively working on a new functionality that will allow the image to be taken without water, which will eliminate this issue.

Presence of Debris

When there are large chunks of debris in the tray, the computer vision may mistakenly identify a piece of debris as a shrimp, resulting in inaccuracies.

To prevent this, simply remove bigger debris from the tray.