FGVC- Aircraft Fine Visual Classification Dataset

The dataset contains 10,200 images of aircraft, with 100 images each of 102 different aircraft model variants, most of which are aircraft. Each image is annotated with a tight bounding box and a hierarchical aircraft model label.

The aircraft model is organized into a four-level hierarchy. The four levels, from fine to coarse, are:

• Model, for example: Boeing 737-76J. As some models are almost visually indistinguishable, this level is not used for evaluation.
• Variants, for example: Boeing 737-700. A variant collapses all visually indistinguishable models into one category. The dataset includes 102 different variants.
• Families, for example: Boeing 737. The dataset includes 70 different families.
• Manufacturer, for example: Boeing. This dataset includes 41 different manufacturers.
  

The data is partitioned into three equally sized subsets of training, validation, and testing. The first two sets can be used for development, and the latter set should only be used for final evaluation. The format of the data is described next.

The creation of this dataset began with a 2012 Johns Hopkins CLSP Summer Workshop on a detailed understanding of objects and scenes in natural images, with the participants, in alphabetical order, including Matthew B. Blaschko, Ross B. Girshick, Juho Kannala, Iasonas Kokkinos, Siddharth Mahendran, Subhransu Maji, Sammy Mohamed, Esa Rahtu, Naomi Saphra, Karen Simonyan, Ben Taskar, Andrea Vedaldi, and David Weiss.