What CANOVA is Applications The software Live Maps Science and models Buy About us
The models
A poster and a photograph are not looked at in the same way. That is why CANOVA has dedicated architectures for the two tasks
I · GRAPHICS
Advertising, packaging, web pages, social posts and applications Communication agencies, freelance graphic artists and designers, artists, illustrators and magazines
Maps at 1 s · 3 s · 7 s Technical details +
II · PHOTOGRAPHY
Select, edit or print photographs Professional photographers, newsrooms, publishers, news outlets and artists
Maps at 1 s · 3 s Technical details +
Visual neural networks simulate the human gaze, but people don't all look the same way.
Two distinct groups of people will never produce two identical attention maps.
That disagreement defines the highest coverage an artificial model can reach.
CANOVA always reports the reference ceiling and the confidence interval. Because accuracy without a reference is not a measurement. CANOVA is not marketing but vision science.
Model coverage Agreement between people
1s
Graphics
0%
Photography
0%
3s
Graphics
0%
Photography
0%
7s
Graphics
0%
Confidence intervals +
0% means a model that does no better than chance 100% a model able to predict visual attention correctly within the agreement between the two test groups of people Be wary of anyone reporting absolute accuracies: they do not exist for these models
Every map is compared against a reference archive built by format and by task: your image is read, analysed and compared with images that are looked at the same way
An analysis without the right benchmark says nothing
Graphics
Over 125,000 artworks
Photography
Over 80,000 photographs
