Stanford Dogs Dataset

The Stanford Dogs dataset contains images of 120 breeds of dogs from around the world. This dataset has been built using images and annotation from ImageNet for the task of fine-grained image categorization. Contents of this dataset:

• Number of categories: 120
• Number of images: 20,580
• Annotations: Class labels, Bounding boxes

• Images (757MB)
• Annotations (21MB)
• Lists, with train/test splits (0.5MB)
• Train Features (1.2GB), Test Features (850MB)
• README

Primary:
  Aditya Khosla, Nityananda Jayadevaprakash, Bangpeng Yao and Li Fei-Fei. Novel dataset for Fine-Grained Image Categorization. First Workshop on Fine-Grained Visual Categorization (FGVC), IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2011.  [pdf]  [poster]  [BibTex]

Secondary:
  J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li and L. Fei-Fei, ImageNet: A Large-Scale Hierarchical Image Database. IEEE Computer Vision and Pattern Recognition (CVPR), 2009.  [pdf]  [BibTex]

Baseline Results This section contains baseline results on two tasks: Mean Accuracy The number of training images per class is varied from 1 to 100. Comparison of Accuracy per Class The accuracy of each class is compared for 15 and 100 training images per class. Experimental Setting All of the experiments use image regions from the bounding box only for both training and testing. The remaining parameters are set to the following values: Type of SIFT descriptors: Grayscale SIFT patch sizes: 8, 10, 14, 18, 22, 26, 30 SIFT grid spacing: 4 pixels Spatial pyramid: 1*1+2*2+4*4 (3 levels) Dictionary Size: 256 Kernel: histogram intersection kernel Contact: aditya86@cs.stanford.edu                  bangpeng@cs.stanford.edu