Attribute-controlled traffic data augmentation using conditional generative models

Amitangshu Mukherjee, Ameya Joshi, Soumik Sarkar, Chinmay Hegde

    Research output: Chapter in Book/Report/Conference proceedingConference contribution

    Abstract

    Perception systems of self-driving vehicles require large amounts of diverse data to be robust against adverse lighting and weather conditions. Collection and annotation of such traffic data is resource-intensive and expensive. To circumvent this challenge, we introduce an approach where we train attribute-based generative models conditioned on the time-of-day labels to reconstruct semantically valid transformed versions of the original data. We further show the generalization capabilities of our model where they are able to reconstruct full traffic scenes despite having only being trained on constrained crops of the original images. Finally, we present a new dataset derived from an original traffic scene dataset augmented with data generated by our attribute-based conditional generative models.

    Original languageEnglish (US)
    Title of host publicationProceedings - 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, CVPRW 2019
    PublisherIEEE Computer Society
    Pages83-87
    Number of pages5
    ISBN (Electronic)9781728125060
    StatePublished - Jun 2019
    Event32nd IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, CVPRW 2019 - Long Beach, United States
    Duration: Jun 16 2019Jun 20 2019

    Publication series

    NameIEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops
    Volume2019-June
    ISSN (Print)2160-7508
    ISSN (Electronic)2160-7516

    Conference

    Conference32nd IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, CVPRW 2019
    Country/TerritoryUnited States
    CityLong Beach
    Period6/16/196/20/19

    ASJC Scopus subject areas

    • Computer Vision and Pattern Recognition
    • Electrical and Electronic Engineering

    Fingerprint

    Dive into the research topics of 'Attribute-controlled traffic data augmentation using conditional generative models'. Together they form a unique fingerprint.

    Cite this