Cool concept and great execution! Just tried a few rounds - the difficulty variation between images keeps it interesting, and the daily format definitely has that Wordle-like habit-forming appeal. I'm curious about a couple of technical aspects: how do you handle the 'ground truth' for scoring accuracy, and what's your image sourcing strategy? Also wondering if you've considered showing how individual guesses compare to the crowd's estimates - that could add another engaging layer to the game.
Hey, thanks for the spot-on questions and the encouragement! You made me remember my initial plan to add a reference to the crowd counting dataset I have used as the ground truth. You can find the link to the dataset in the About page (the ? icon) now. These datasets typically include crowd annotations every image where every person is annotated with head locations (and sometimes the head size) so rendering that was trivial. I actually do show the accuracy of the individual guesses at the end (with the summary of every round) and during the game (with that animated accuracy ruler). It would be great to find out if they are not working properly or not explanatory enough.
1 comments