The MURDOC project introduces an application to enhance trustworthiness and explainability in computer vision models, focusing on camouflage detection. It aims to address the need for transparent and interpretable AI systems in sensitive domains such as security and defense. MURDOC integrates advanced eXplainable AI techniques including off-ramps for collecting interpretable insights, attention mechanisms for high-lighting relevant features, and a user-centric image pre-processing tool within the visualization interface. This paper offers an assessment of MURDOC's potential impact on trustworthiness and explainability in camouflaged object detection tasks. It discusses and assesses the potential effectiveness of off-ramps in gathering XAI output at various model stages, such as feature maps, attention mechanisms, and activation maps. Additionally, the paper investigates how user-driven image pre-processing mechanisms may enhance the trustworthiness of the model's predictions and decisions, allowing users to modify the input and observe prediction changes. As a work in progress, the development of MURDOC shows promise in bridging the gap for transparency and interpretability in camouflage detection. The paper also discusses the challenges, future directions, and potential applications of MURDOC's capabilities in diverse domains requiring transparent and trustworthy AI systems. The source code for the version of MURDOC in this paper is available on GitHub at: https://github.com/Department-70/DASC_2024.


    Zugriff

    Zugriff prüfen

    Verfügbarkeit in meiner Bibliothek prüfen

    Bestellung bei Subito €


    Exportieren, teilen und zitieren



    Titel :

    MURDOC: Transforming Pixels into Perception for Camouflage Detection


    Beteiligte:
    Hogue, Debra (Autor:in) / Kastl, Zak (Autor:in) / Karch, Joseph (Autor:in) / Nguyen, Don (Autor:in) / Schley, Lacey (Autor:in) / Lewis, Justin (Autor:in) / Connelly, Shane (Autor:in) / Weaver, Chris (Autor:in)


    Erscheinungsdatum :

    29.09.2024


    Format / Umfang :

    2382417 byte





    Medientyp :

    Aufsatz (Konferenz)


    Format :

    Elektronische Ressource


    Sprache :

    Englisch



    Camouflage assessment considering human perception data

    Gretzmacher, F.M. / Ruppert, G.S. / Nyberg, S. | Tema Archiv | 1998


    Camouflage assessment considering human perception data [3375-08]

    Gretzmacher, F. M. / Ruppert, G. S. / Nyberg, S. et al. | British Library Conference Proceedings | 1998


    Camouflage

    Goodwin, M. | Engineering Index Backfile | 1944


    Radar Camouflage

    L. Lewin | NTIS | 1945


    Ship camouflage

    Blodgett, Leo S | DSpace@MIT | 1919

    Freier Zugriff