You see it? You name it!

The preliminary results are here!

Thank you very much to the +3700 of you that helped us getting here! 🙏 The first batch of 4k images has been labeled and classified 🙌 We will soon be uploading 6k more images for you to have fun with, but first, let's have a look together at some preliminary data Here is a figure that summarizes some of the current results 👇 Please note that we wish NOT to bias (too much) your future labeling/classifying work.

First of all, the top row tell us that the vast majority of images are consistently assigned to the same category, irrespective of whether we consider the first break of the classification tree (living/animate vs. nonliving/inanimate), the second (human vs. animal, natural vs. artificial) or the third (the specific category). However, when we reach this level (the last graph on the right of the top row), there is clearly increase variance: it is (a bit) harder to find an agreement. Specifically, some images are easy (in light blue: a musical instrument, according to ALL of you), while other are harder (in green, something that is hard to tell whether is a decoration, a tool, an hygiene item, etc...).

Then, we can look at the actual labels provided for each image. As you know there is no right or wrong answer, yet we confess we did have expected labels in mind. And we can check whether the ones you provided fall within what we expect or not (see the first graph on the bottom row). Note how this value can be negative as for some images not only you ALL provided the expected label, but you also included many plausible synonyms that fall within our range of predicted answers (basically, virtually everyone said both turtle and tortoise for the orange image). On the other side of the spectrum, are images for which you provided (quite a few) labels we were not necessarily expecting: for instance, for the image in red we'd expect women, girl, female face, etc.. but you surprised us with graduation, college graduate, etc... which are all obviously correct! See? We need you & there are no wrong choices! 😆 We can also look at the overall spread of responses (or entropy of responses, right graph on the bottom row): an index of how many different labels where provided for the same image. Again, for some images, there are really no other options (a peacock is peacock, image in light purple), while for others we could be discussing on what they are and how to call them for...a while! What is that object in dark blue? You provided a pretty wide range of options 😅

Please do note that these are just representative examples of partial results: any conclusion will be drawn (and shared1) once all the images have been labeled. In the meantime, feel free to ask us questions in the talk boards.

---

And in general remember that...

Once we will have proper labels and classification for all the images, we will be releasing the final dataset on our main portal.

We will also be sharing the result of all neuroimaging and machine learning studies that will be using this dataset. To be updated on these outcomes, don't miss our publications.

Finally, you might find talks related to this project on our YouTube channel and frequent updates on our research on our Twitter account. For instance, at this link you can find a playlist with our latest accomplishments.