Section: New Results

Fine-Grained models of objects and texture

Paticipants: Iasonas Kokkinos, Matthew Blaschko, Stavros Tsogkas, Andrea Vedaldi, Mircea Cimpol, Subhransu Maji, Ross Girshick, Juho Kannala, Esa Rahtu, David Weiss, Ben Taskar, Karen Simonyan.

In [31] and [22] we explore methods for the fine-grained understanding of objects and textures, respectively.

In [22] we introduce a texture dataset that is accompanied by descriptions that capture the essence of the textures in terms of attributes. We explore a broad range of classification techniques for these texture attributes and demonstrate that the learned classifiers help improve generic texture recognition methods.

In [31] we introduce a large-scale dataset of airplanes that is accompanied by thorough human annotations at different levels: airplane types, segment lineouts, attributes, and part descriptions are provided for more than 7000 airplane images. We explore the potential of these rich annotations for the task of constructed fine-grained image descriptions using discriminative training techniques on top of standard image representations (Histogram-of-gradient features).