A discussion forum for the Food Recognition Challenge on AIcrowd.
I took a quick look at the dataset and I feel it lacks quality.
Many segmentations are not very precise (see example below for cheese). I think that the algorithm that does the best job, given the low quality of the data, will not really have the highest IoU…
Hi @gloria_macia_munoz - some of the segmentations, as the one you identified, have been done by humans under time constraints. Those can certainly be improved. As we get better and better algorithms, more and more high quality data will become available for this challenge, in future rounds. It’s very important to note that this will be a long-term challenge / benchmark. This is just the first round, with the data that was provided. Over time, both the quantity and the quality will increase.