I came across this amazing tool "ArtEmis", which accounts for subjectivity and variability in human response.

I have been working recently to develop a tool/ interface to collect subjective responses to images in my own archive. After collecting those responses, I intend to compare the variability of those responses. The more the variation, the more difficult it is to train AI for such images.

Interestingly, I came across the tool while asking chat-gpt about it. It mentioned about artemis which I checked to know that it's a project by a scientist in stanford. His name is Panos Achlioptas.

References:

https://hai.stanford.edu/news/artists-intent-ai-recognizes-emotions-visual-art

Link to the research paper by Panos: https://arxiv.org/pdf/2101.07396

Website: https://affective-explanations.org