Not logged in. Login | Signup


Download Features
SIFT features

We currently provide densely sampled SIFT[1] features. We provide raw SIFT descriptors as well as quantized codewords. Spatial coordiates of each descriptor/codeword are also included. The quantized codewords are suitable for Bag of Words representations[2][3]. The features are packaged as Matlab files and can be freely downloaded ( no signing-in is required ). Details are as follows:

  • Each image is resized to have a max side length of no more than 300 pixel. SIFT descriptors are computed on 20x20 overlapping patches with a spacing of 10 pixels. Images are further downsized (to 1/2 the side length and then 1/4 of the side length) and more descriptors are computed. We use the VLFeat[4] implemenation of dense SIFT (version
  • We perform k-means clustering of a random subset of 10 million SIFT descriptors to form a visual vocabulary of 1000 visual words. Each SIFT descriptor is quantized into a visual word using the nearest cluster center.
    1. David G. Lowe, Distinctive Image Features from Scale-Invariant Keypoints. International Journal of Computer Vision, 2004. pdf
    2. L. Fei-Fei and P. Perona, A Bayesian Hierarchical Model for Learning Natural Scene Categories. IEEE Comp. Vis. Patt. Recog. 2005. pdf
    3. Svetlana Lazebnik, Cordelia Schmid and Jean Ponce, Beyond Bags of Features: Spatial Pyramid Matching for Recognizing Natural Scene Categories. IEEE Comp. Vis. Patt. Recog. 2006. pdf
    4. A. Vedaldi and B. Fulkerson. VLFeat: An Open and Portable Library of Computer Vision Algorithms. 2008.
    How to download?
    1. We have not yet released SIFT features for all synsets. To check the list of synsets with SIFT features released, please use the API:

      You can click here to obtain the synset names.

    2. When you browse ImageNet from the Explore page, you can download the bag of visual words (sbow) feature of a synset if there is an icon "Download BoW Feature" below the image view panel.
    3. You can download the bag of visual words ( sbow ) feature for a given synset using the API:

      The API will return a Matlab ( .mat ) file. In the Matlab file, each descriptor has 5 fields: x, y, norm, scale, word. The word field is the index of the cluster center, i.e. an integer between 0 and 999.
    Code for computing the features
    To learn more about downloading using the HTTP protocol, please refer to API documentation.