ETHZ Shape Classes

Vittorio FerrariTinne Tuytelaars, Luc Van Gool


This dataset contains images of five diverse shape-based classes, collected from Flickr and Google Images. The main challenges it offers are clutter, intra-class shape variability, and scale changes. We deliberately selected several images where the object comprises only a rather small portion of the image, and we made an effort to include objects appearing at a wide range of scales. The objects are mostly unoccluded and are all taken from approximately the same viewpoint (the side).

In [1] we tackled the challenge of detecting objects in real images given a single hand-drawn example as ‘model’. If you would like to address the same problem,
the hand-drawings are included in this release. We recommend using all 255 images as test set for every class. Hence, to search for, e.g. bottles also in images images of mugs, swans, etc., making for a large negative test set. This is important as it allows to get a reliable value for the incidence of false-positives generated by the detection algorithm.

In addition to its use in the above context, this dataset is also suited for the conventional setting in which models are learnt from real images (for example, by splitting the dataset in half training / half testing). We have released further results in this setting in [2,3,4]. Moreover [3,4] also report experiments in the setting of [1], i.e. using a single hand-drawn example as a model.

Alternative annotations by Xavier Giro:
You can download binary mask and alternative bounding-box annotations by Xavier Giro. These are useful for algorithms requiring a solid mask as input or for evaluation, rather than a contour. Important:this material is not part of the official protocol. If you want to compare to previous works, please use the annotations in the main package above.


Apple Logo Bottle Giraffe Mug Swan


Apple Logos
(40 images)
(48 images)
(87 images)
(48 images)
(32 images)


Filename Description Size
ethz_shape_classes_v12.tgz The entire ETHZ Shape Class Dataset 30MB


V 1.2

  • added ground-truth outlines
  • added performance plots for [1,2,3,4]
  • clarified README.txt

V 1.01

  • fixed annotation for the Cuvee bottle (used to be a single pixel).
    Thanks to Nhon Trinh and Mario Fritz for reporting this bug !

V 1.0

  • initial release


  1. Object Detection by Contour Segment Networks
    Vittorio Ferrari, Tinne Tuytelaars and Luc Van Gool,
    In European Conference in Computer Vision (ECCV), 2006.
  2. Groups of Adjacent Segments for Object Detection
    Vittorio Ferrari, Frederic Jurie, and Cordelia Schmid,
    In IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), 2008.
  3. Accurate Object Detection with Deformable Shape Models Learnt from Images
    V. Ferrari, F. Jurie, and C. Schmid,
    In Computer Vision and Pattern Recognition (CVPR), 2007.
  4. From Images to Shape Models for Object Detection
    V. Ferrari, F. Jurie, and C. Schmid,
    In International Journal of Computer Vision (IJCV), 2010.