This is my customary update of percentage of deep-learning papers in the major vision conferences. At CVPR 2015, I described its exponential growth, and after ICCV 2015 I wondered whether it had plateaued.
Since the CVF has not released the papers yet, I had to adapt my script to take the titles from the program.
[UPDATE] CVF papers are online, so we can use the generic script shown below.
You can download all titles in TXT format in here if you want to play around. Here the evolution of the results:
And the plot using XKCD style, as described in this blog post:
So the plateau was in fact only a short pit stop to take over again. Maybe it saturated in the classical areas such as object detection or image labeling and is now expanding to the rest of the topics?
Here the updated code for you to try:
And here the old code to parse it from the program: