Training machines to see scenic beauty
summary
Claims a machine can predict scenic beauty after training on 215,000 images labeled via 1.5 million ratings using a convolutional neural network with transfer learning. Results show both natural and built features matter, horizon visibility boosts scores, flat grass scores low, and models map beauty across Great Britain and 200,000 London images. Outputs seed a nonprofit and an on‑device mobile app that personalizes suggestions; reported experiments and maps indicate the approach works at city and national scales.
AI-written summary · about
chapters (8)
- 0:00AI and beauty intro
- 1:05PhD beauty and wellbeing
- 3:12Beauty perception theories
- 4:15Deep learning for beauty
- 6:02Training beauty detection
- 7:27AI beauty map of London
- 8:53Beau Places with AI
- 10:39AI beauty future
related talks
39mInsights About Places with Deep Learning Computer Vision
40mMachines are Learning
32mTeach JS About Aesthetics with Machine Learning
33mRunning Machine Learning in Production - a Journey to Success
33mWhat and where? - Machine learning for geospatial image analysis
33mMachine Learning in Visual Search
41mPractical Aspects of Applying Modern Computer Vision Technologies
30mMachine learning applications for rock art research