Hi @Amy_Bray
Thanks for the update on geometry usage. Since some earlier submissions may have used geometrical features, could you please update the leaderboard to reflect only image-based models, as per the current rules?
This will help ensure fairness and accurate evaluation.
Yes, I thought of the same, now they will need to change the test data. I believe most of the top scores used location features.
I agree that the LB should be reset, but will refreshing alone help?
It will only help if the new test ids have been shuffled , people can decide to just use the old data no?
Exactly my thoughts too. This competition's rules keep changing, they need to settle on one thing now. Makes me think they did not plan it well.
@Koleshjr, reset or no reset, I don't think that will help for now. Based on my understanding, I believe @Zindi allows us to extract additional images for the train dataset while keeping the test data images fixed, as provided in the s2_images.zip file. However, I would still like @Zindi to confirm this, because even if we are extracting images for the train data, we cannot proceed without using the 'trainGeo.geojson' file. Unfortunately, this file cannot be downloaded, despite being provided. Because of this, we are limited to using only the datasets provided, excluding the 'trainGeo.geojson' file, without any additional image extraction.
I think a reset is surely needed, Otherwise then the competition will be unfair since the actual "results" from the new update then won't be seen which helps in motivating people to do better? @MICADEE
@ yeah... well said... Let's wait for @Zindi to do the needful on this.
Of course @Koleshjr I agreed, a reset would make it fair and give everyone a clear view of progress with the update.
@Zindi and @Amy_bray