Exploring the Bounds of the Utility of Context for Object Detection

Abstract

The recurring context in which objects appear holds valuable information that can be employed to predict their existence. This intuitive observation indeed led many researchers to endow appearance-based detectors with explicit reasoning about context. The underlying thesis suggests that stronger contextual relations would facilitate greater improvements in detection capacity. In practice, however, the observed improvement in many cases is modest at best, and often only marginal. In this work we seek to improve our understanding of this phenomenon, in part by pursuing an opposite approach. Instead of attempting to improve detection scores by employing context, we treat the utility of context as an optimization problem: to what extent can detection scores be improved by considering context or any other kind of additional information? With this approach we explore the bounds on improvement by using contextual relations between objects and provide a tool for identifying the most helpful ones. We show that simple cooccurrence relations can often provide large gains, while in other cases a significant improvement is simply impossible or impractical with either co-occurrence or more precise spatial relations. To better understand these results we then analyze the ability of context to handle different types of false detections, revealing that tested contextual information cannot ameliorate localization errors, severely limiting its gains. These and additional insights further our understanding on where and why utilization of context for object detection succeeds and fails.

Exploiting Multi-Object Relationships for Detecting Adversarial Attacks in Complex Scenes

“…YOLO takes a single pass design aiming to reducing the computation complexity and improving detection speed. Context-aware Object Detection aims to exploit context information to boost the performance of object detection [34,39,8,2]. Earlier approaches incorporate context (object co-occurrence) information as a post-processing step to re-score objects detected by DNN-based object detectors [12,6,33].…”

Role of Spatial Context in Adversarial Robustness for Object Detection

“…[12] learns to use stuff in the scene to find objects. [22,2,27] discuss the role of context in object detection while [29] for classification and segmentation. [40] shows that a network trained for scene classification has implicitly learned object detection which suggests the inherent connection between scene classification and object detection.…”

“…We believe (1) can be achieved without exploiting the contextual reasoning by generating lots of false positives at the patch location to dominate the mAP calculation, which may not necessarily affect the true positives. However, since we are interested in showing the exploitation of contextual reasoning, we focus on (2) where the adversary should make the detector blind by reducing true positives.…”

“…The computer vision community has studied the role of spatial context in object detection for a long time. It is wellknown that object detection performance is improved by considering information from the entire scene [6,36,5,2]. Fast object detectors (YOLO [25], SSD [17], and Faster-RCNN [26]) process the entire image in one forward pass to reduce inference time and hence implicitly use contextual information for their predictions.…”

Hallucination In Object Detection — A Study In Visual Part VERIFICATION

“…Here, we propose to evaluate visual verification not with precision but with a cost-weighted variant of recall. Object hallucination by deep detectors can be causes by sensitivity to the absolute position in the image [14,15] while also affected by scene context [16,17,18,19,20]. Here, we focus on the visual verification task, its evaluation measure, a novel dataset, and a comparison of popular existing detectors.…”