Visual and Linguistic Treebank (VLT2K)
Release 3, 8th January 2015
The Visual and Linguistic Treebank contains multiple descriptions for the 2,424 images in the trainval portion of the PASCAL VOC 2010 Action Recognition Taster. There is also object annotations for 431 images, and corresponding Visual Dependency Representations for the object-annotated images.
The original images can be downloaded directly from PASCAL at http://pascallin.ecs.soton.ac.uk/challenges/VOC/voc2010/#devkit
Please direct any comments or problems to: firstname.lastname@example.org
An overview of the Visual and Linguistic Treebank Dataset.
The files in this directory will be the JPEGs of the images for which descriptions and annotations are available. You need to download the JPEGs directly from PASCAL.
The files in this directory are the object boundary annotations created from LabelMe annotations. There is one .xml file per annotated image.
The VDR annotations for the images. -1,-2,-3 suffixes mean they correspond to the 1st, 2nd, and 3rd description respectively.
.actions is the list of actions in the image
We collected three descriptions for each image from Amazon Mechanical Turk. The -1,-2,-3 suffixes mean they correspond to the 1st, 2nd, and 3rd description respectively.
.desc is the raw written description as collected from Mechanical Turk