Description

We collected additional data for the PARSE dataset for the ECCV 2014 paper, "Zero-Shot Learning via Visual Abstraction." To supplement the original pose annotation, we collected facial expression, gaze direction, and gender for the PARSE images. We also grouped 108 of the 305 images into 14 sematic pose categories, such as "The person is pitching a ball," "The person is catching a ball in a baseball game," "The person is celebrating a goal/point," and "The person is batting during a sports game (e.g., cricket, baseball)."

Download PARSE Dataset Annotations


In addition to the PARSE dataset annotations, we also collected 696 illustrations depicting the same 14 categories as PARSE. Identical annotations are also available for these illustrations.

Broken path. :( Broken path. :( Broken path. :( Broken path. :(

View Abstract-Category-Level-PARSE Dataset


We also collected 914 illustrations where people tried to re-create the PARSE real images. There are usually 2-3 illustrations for each real image. Since the users only had a 2 seconds to see the real image, the illustrations vary from the real images, particularly for the less salient/important details. Identical annotations are also available for these illustrations.

Broken path. :(
Broken path. :(
Broken path. :(
Broken path. :(
Broken path. :(
Broken path. :(
Broken path. :(
Broken path. :(

View Abstract-Instance-Level-PARSE


Downloads

PARSE Image Dataset: [Annotations] [README]

Abstract-Category-Level-PARSE Illustration Dataset: [Images] [Annotations] [README]

Abstract-Instance-Level-PARSE Illustration Dataset: [Images] [Annotations] [README]