You can follow this topic https://zindi.africa/competitions/barbados-lands-and-surveys-plot-automation-challenge/discussions/28640

It seems we won’t receive any response from Zindi. Therefore, I recommend not using manual labeling. You can take the risk, but the chance of being disqualified from the leaderboard is quite high. I’m also developing segmentation models using only the noisy labels from the competition and can achieve around 0.95 IoU. If manual label correction were allowed, I think it could reach about 0.97, but I prefer not to take that risk. Good luck!