Dear Zindians,

This is a double whammy. There are two solutions.

Winning solutions for the crop yield: https://github.com/ZindiAfrica/Computer-Vision/tree/main/Image%20Classification/CGIAR%20Crop%20Yield%20Prediction%20Challenge

When setting up this competition the competition hosts were aware that the data might be misleading but they needed the solution so they went ahead with the competition. The problem with the data is that farmers had to indicate where their fields were but sometimes farmers were a bit off when indicating their fields and it was impossible to indicate the center of the field as it sometimes meant walking into the middle of the field at full growth.

Thus a second competition was born from this, the Lacuna - Correct Field Detection Challenge. This competition was to take the farmer-indicated geolocations and correct them to the center of the field.

Here are the solutions of this comp: https://github.com/ZindiAfrica/Computer-Vision/tree/main/Image%20Classification/Lacuna%20Correct%20Field%20Detection%20Challenge

If you were to apply the field correction algorithm and then the yield estimation algorithm does your result improve?

We'd love to hear your experience try this out!