Dropout Sampling for Robust Object Detection in Open-Set Conditions

By Dimity Miller, Lachlan Nicholson, Feras Dayoub, Niko Sünderhauf. 2017

272 citations recorded by the source corpus.

Dropout Variational Inference, or Dropout Sampling, has been recently proposed as an approximation technique for Bayesian Deep Learning and evaluated for image classification and regression tasks. This paper investigates the utility of Dropout Sampling for object detection for the first time. We demonstrate how label uncertainty can be extracted from a state-of-the-art object detection system via Dropout Sampling. We evaluate this approach on a large synthetic dataset of 30,000 images, and a real-world dataset captured by a mobile robot in a versatile campus environment. We show that this uncertainty can be utilized to increase object detection performance under the open-set conditions that are typically encountered in robotic vision. A Dropout Sampling network is shown to achieve a 12.3 % increase in recall (for the same precision score as a standard network) and a 15.1 % increase in precision (for the same recall score as the standard network).

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

Citations in the corpus

Towards Open World Object Detection (2021, methodology): “However, we ﬁnd that increasing N ex to large values does-not help and at the same time adversely affect how unknowns are handled (evident from WI and A-OSE scores).”
Towards Safe Autonomous Driving: Capture Uncertainty in the Deep Neural Network For Lidar 3D Vehicle Detection (2018, methodology): “This dropout sampling method has been applied to active learning for cancer diagnosis [25], semantic segmentation [26], [27] and image object detection for openset problem [28].”
The limits and potentials of deep learning for robotics (2018, methodology): “One way to handle the open-set problem and identify unknowns is to utilize the epistemic uncertainty [28], [54] of the model predictions to reject predictions with low conﬁdence [76].”
A Wholistic View of Continual Learning with Deep Neural Networks: Forgotten Lessons and the Bridge to Active and Open World Learning (2020, background): “In deep networks this could be achieved by assessing the variations of stochastic forward passes through a neural network with dropout [38], [146], [147], as a variational Bayesian approximation to a distribution on the weights [37], or by empirically estimating the output’s variability with respect”
Evaluating Merging Strategies for Sampling-based Uncertainty Techniques in Object Detection (2018, methodology): “The clustering method used in [8] (BSAS IoU 0.”
A Review and Comparative Study on Probabilistic Object Detection in Autonomous Driving (2020, methodology): “Finally, [6], [7], [87] take the beneﬁts of redundant detection outputs to compute sample statistics of probability distributions, instead of directly-modeling the output probability and discarding redundant detections.”
The Overlooked Elephant of Object Detection: Open Set (2020, methodology): “[19] also attempted to approach object detection as an open-set problem, they did not provide a formalization of the problem and its deep impact on the applicability of object detectors in the real world.”
OW-DETR: Open-world Detection Transformer (2021, methodology): “We follow the same evaluation protocol of [23] and report the performance in Tab.”
VOS: Learning What You Don't Know by Virtual Outlier Synthesis (2022, methodology): “Several works (Dhamija et al., 2020; Miller et al., 2019; 2018; Hall et al., 2020; Deepshikha et al., 2021) used approximate Bayesian methods, such as MC-Dropout (Gal & Ghahramani, 2016) for OOD detection.”
QuadricSLAM: Dual Quadrics From Object Detections as Landmarks in Object-Oriented SLAM (2018, methodology): “as proposed by [30]) into the SLAM system via the covariance matrices Λij in future work.”
Opening up Open World Tracking (2021, background): “Object detection has recently been studied in open-set conditions [20, 51].”
Probabilistic Object Detection: Definition and Evaluation (2018, methodology): “Evaluated Detectors The state-of-the-art conventional object detectors evaluated were SSD-300 [27], YOLOv3 [37], FasterRCNN with ResNet backbone (FRCNN R) [51], FasterRCNN with ResNet backbone and feature pyramid network (FPN) (FRCNN R+FPN) [28], and FasterRCNN with ResNeXt backbone and FPN (FRCNN X”
Deep Multi-Modal Object Detection and Semantic Segmentation for Autonomous Driving: Datasets, Methods, and Challenges (2019, methodology): “It has been applied to semantic segmentation [207] and open-world object detection problems [208],”