TranSketch Dataset: Learning to Transform Sketches

Luming Chen

EECS Department, University of California, Berkeley

Technical Report No. UCB/EECS-2020-92

May 29, 2020

This publication is archived. It is kept only for reference purposes, so it is no longer being updated and may not meet accessibility standards. If you need this content in a different format, please email webteam@eecs.berkeley.edu.

http://www2.eecs.berkeley.edu/Pubs/TechRpts/2020/Archive/EECS-2020-92.pdf

This report introduces a new dataset TranSketch, which is a collection of 20,032 tree sketch pairs, each accompanied with a natural language description describing the instructions needed to transform one sketch to the other. To provide more options for exploration, for each sketch, we also include a latent representation produced by the encoder of a Sketch-RNN model. The TranSketch dataset provides fine-grain transformation between stroke-based sketch pairs using stylistic content text descriptions. Our statistical analysis suggests tree component-level transformation to be the most promising direction for investigation, but there are certainly more to be explored for this dataset.

Advisors: John F. Canny

BibTeX citation:

@mastersthesis{Chen:EECS-2020-92,
    Author= {Chen, Luming},
    Title= {TranSketch Dataset: Learning to Transform Sketches},
    School= {EECS Department, University of California, Berkeley},
    Year= {2020},
    Month= {May},
    Url= {http://www2.eecs.berkeley.edu/Pubs/TechRpts/2020/EECS-2020-92.html},
    Number= {UCB/EECS-2020-92},
    Abstract= {This report introduces a new dataset TranSketch, which is a collection of 20,032 tree sketch pairs, each accompanied with a natural language description describing the instructions needed to transform one sketch to the other. To provide more options for exploration, for each sketch, we also include a latent representation produced by the encoder of a Sketch-RNN model. The TranSketch dataset provides fine-grain transformation between stroke-based sketch pairs using stylistic content text descriptions. Our statistical analysis suggests tree component-level transformation to be the most promising direction for investigation, but there are certainly more to be explored for this dataset.},
}

EndNote citation:

%0 Thesis
%A Chen, Luming 
%T TranSketch Dataset: Learning to Transform Sketches
%I EECS Department, University of California, Berkeley
%D 2020
%8 May 29
%@ UCB/EECS-2020-92
%U http://www2.eecs.berkeley.edu/Pubs/TechRpts/2020/EECS-2020-92.html
%F Chen:EECS-2020-92