Hi Organizers,

First, thank you for hosting a very interesting competition! I have learned a lot about many different topics, such as acquiring satellite data and understanding EPSG codes for projections.

@meganomaly mentioned at the beginning of the competition that selected submissions from the top 10 would be scored. Since this competition is based on accuracy as a metric, I experimented by submitting submissions containing only zeros and ones. I observed that the metric treats all non-submitted rows as incorrect. This implies that, for example, the top scorer with a score of 0.979390829 must have submitted at least 97.9% of all the rows. However, it was stated that we should only submit a few rows at this stage. I submitted about a million rows, but this makes it impossible to achieve such a high score under the current guidelines.

I used an attention-based segmentation model and did a lot preprocessing of data retrieved from the coordinates of the runways. According to my evals, my model has good performance, but I am unable to achieve a higher score due to the aforementioned issue.

Given this situation, how will you ensure that you run the full scoring on the correct submissions?

Thanks for reading and your assistance!

Best regards, Emil