Abstract:
- Overview 256 Object Categories + Clutter At least 80 images per category 30608 images instead of 9144
- Caltech-101: Drawbacks Smallest category size is 31 images: Too easy? left-right aligned Rotation artifacts Soon will saturate performance
-
Caltech-256 : New Features
Smallest category size now 80 images Harder Not left-right aligned No artifacts Performance is halved More categories New and larger clutter category
-
Collection Procedure Similar to Caltech-101 (Li, Fergus, Perona)
Four sorters rate the images 1 good: a clear example 2 bad: confusing, occluded, cluttered, or artistic 3 not applicable: object category not present
92,652 Images from Google and Picsearch 32.1% were rated good and kept
Some images borrowed from 29 of the largest Caltech-101 categories (green)
- Acknowledgements Rob Fergus and Fei Fei Li, Pierre Moreels for code and procedures developed for the Caltech-101 image set Marco Ranzato and Claudio Fanti for miscellaneous help Sorters: Lis Fano, Nick Lo, Julie May, Weiyu Xu for making this image set possible with their hard work
Please site as: Griffin, G. Holub, AD. Perona, P. The Caltech 256. Caltech Technical Report. The technical report will be available shortly.
Reference:
