PlatformMarketplaceSolutionsResourcesOpen DatasetsCommunityCompany
Animal Parts
Jun 20, 2021 9:58 AM


This page describes the Animal Parts dataset - annotations of animal parts in ILSVRC2012 images.The dataset contains eye and foot keypoint annotations for 15K images of 100 animal classes from the "vertebrate" subtree of ILSVRC2012.The dataset provides only keypoint and bounding box annotations. For the actual images please visit the ILSVRC2012 homepage and download the corresponding images from there.

Data Format

imdb-animalParts-eye.mat; imdb-animalParts-foot.mat

  • eye/foot annotations in compact MATLAB imdb structures
    • structure reference:
      • imdb.images - image database
      • - image filename
      • imdb.images.set - image set (1-train,2-test)
      • - unique id of an image - identifies bounding boxes present in the image (imdb.bbx.imageId)
    • imdb.bbx - bounding box database
      • imdb.bbx.imageId - points to the image id containing a given bounding box
      • imdb.bbx.class - animal class of a given box
      • - unique id of a box - identifies keypoints peresent inside the bounding box
      • imdb.bbx.set - box set (1-train,2-test)
      • imdb.bbx.loc - location of the box (xmin,ymin,xmax,ymax)
    • - keypoint annotations
      • - points to the box id containing a given keypoint
      • - location of the keypoint (x,y)
      • - corresponding animal class
      • - a binary flag indicating whether a keypoint cannot be uniquely assigned to one bounding box
        • images containing ambiguous keypoints should be skipped during evaluation!
    • - keypoint class - contains only "1" for now
      • imdb.classes - information about animal classes
      • - wnid of each class
      • imdb.classes.description - imagenet description of each class
      • imdb.classes.ok - the classes that have available annotaions
      • imdb.classes.source - source set of classes as used in [1]
      • - target set of classes as used in [1]

xml/eye; xml/foot

  • annotations in the pascal xml format
  • individual xml tags, mostly self explanatory
  • name: imagenet wnid of the given bounding box animal class
  • description: imagenet description of the given bounding box animal class
  • classId: id of the given bounding box animal class
  • please double-check that after parsing the xml files, your dataset is identical to the attached matlab imdb structures

sourceClasses.txt; targetClasses.txt

  • class ids of the source and target classes as used in [1]

classMapDescription.txt; classMapWNID.txt

  • the mapping between the original ImageNet classes and the ones used in [1]

    • classMapWNID.txt - the mapping between WNIDS
  • classMapDescription.txt - the mapping between descriptions of WNIDS (the lines between the two files correspond)

    • each line starts with a binary flag indicating whether the original class was altered followed by a pair "original imagenet class", "new class"


Please use the following citation when referencing the dataset:

     author = {D. Novotny and D. Larlus and A. Vedaldi},
     booktitle = {Proceedings of the British Machine Vision Conference (BMVC)},
     title = {I Have Seen Enough: Transferring Parts Across Categories},
     year = {2016}
🎉Many thanks to Graviti Open Datasets for contributing the dataset
Basic Information
Application ScenariosNot Available
AnnotationsNot Available
TasksNot Available
Updated on2021-01-20 03:11:50
Data TypeNot Available
Data Volume0
Annotation Amount0
File Size0.00B
Copyright Owner
VGG of University of Oxford