Detection-by-Localization: Maintenance-Free Change Object Detector

Abstract

Recent researches demonstrate that self-localization performance is a very useful measure of likelihood-of-change (LoC) for change detection. In this paper, this "detection-by-localization" scheme is studied in a novel generalized task of object-level change detection. In our framework, a given query image is segmented into object-level subimages (termed "scene parts"), which are then converted to subimage-level pixel-wise LoC maps via the detection-by-localization scheme. Our approach models a self-localization system as a ranking function, outputting a ranked list of reference images, without requiring relevance score. Thanks to this new setting, we can generalize our approach to a broad class of self-localization systems. Our ranking based self-localization model allows to fuse self-localization results from different modalities via an unsupervised rank fusion derived from a field of multi-modal information retrieval (MMR). This study is a first step towards a maintenance-free approach, minimizing maintenance cost of map maintenance system (e.g., background model, detector engine).We would like to express our sincere gratitude to Takuma Sugimoto, Rino Ide and Kousuke Yamaguchi for development of deep learning architecture, and initial investigation on change detection tasks on the dataset, which helped us to focus on our Detection-by-Localization project.

Diagnosing Deep Self-localization Network: Accelerated Model-free Subimage-level Diagnosis via Image Synthesis and Relevance Feedback

“…In a subimage-level diagnosis framework, one pixel may belong to multiple overlapping RoI regions, and thus receive multiple different IR values from such overlapping RoIs. Here, we fuse such multiple IRs into a single IR value, with a similar information fusion method as in our previous work in [2], which is theoretically supported by multi-modal information retrieval in [9]. That is, we start with a set of multiple diagnosis results in the form of multiple PDF vectors.…”

“…Such a DSL technique enables a DA module to focus the available resources on the domain-shifted region, rather than the entire workspace, which leads to significant reduction in the perdomain DA cost for data collection and retraining. In our approach, the DSL task is formulated as a fault-diagnosis (FD) problem as in our previous study [2], in which the deterioration of the DCN classifier for a given query image is viewed as an indicator of domain-shifts at the imaged region. Figure 1 shows an overview of the DSL task for a typical self-localization system based on 3D point cloud imagery.…”