Is Faster R-CNN Doing Well for Pedestrian Detection?

By Liliang Zhang, Liang Lin, Xiaodan Liang, Kaiming He. 2016

836 citations recorded by the source corpus.

Detecting pedestrian has been arguably addressed as a special topic beyond general object detection. Although recent deep learning object detectors such as Fast/Faster R-CNN have shown excellent performance for general object detection, they have limited success for detecting pedestrian, and previous leading pedestrian detectors were in general hybrid methods combining hand-crafted and deep convolutional features. In this paper, we investigate issues involving Faster R-CNN for pedestrian detection. We discover that the Region Proposal Network (RPN) in Faster R-CNN indeed performs well as a stand-alone pedestrian detector, but surprisingly, the downstream classifier degrades the results. We argue that two reasons account for the unsatisfactory accuracy: (i) insufficient resolution of feature maps for handling small instances, and (ii) lack of any bootstrapping strategy for mining hard negative examples. Driven by these observations, we propose a very simple but effective baseline for pedestrian detection, using an RPN followed by boosted forests on shared, high-resolution convolutional feature maps. We comprehensively evaluate this method on several benchmarks (Caltech, INRIA, ETH, and KITTI), presenting competitive accuracy and good speed. Code will be made publicly available.

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

SSD: Single Shot MultiBox Detector (2015, methodology): “Besides, since our predictions do not rely on ROI pooling as in [6], we do not have the collapsing bins problem in low-resolution feature maps [22].”
Domain Adaptive Faster R-CNN for Object Detection in the Wild (2018, methodology): “Readers are referred to [60] for more details about the architecture and the training procedure.”
Computer vision and deep learning techniques for pedestrian detection and tracking: A survey (2018, methodology): “Automatic features extraction in also used in the work by Zhang et al., where the authors used faster r-cnn for pedestrian detection [190] .”
Recurrent Topic-Transition GAN for Visual Paragraph Generation (2017, background): “Recently, great advances [18, 3, 4, 31, 24, 33, 22, 21, 36] have been achieved in describing images and videos using a single high-level sentence, owing to the advent of large datasets [23, 17] pairing images with natural language descriptions.”
Deep Learning for Generic Object Detection: A Survey (2018, neutral): “IEEE SPL 23(10):1499–1503 2 [307] Zhang L.”
Improving Multispectral Pedestrian Detection by Addressing Modality Imbalance Problems (2020, methodology): “Inspired by [41], Boosted Decision Trees classifier [21] is built on high-resolution RPN feature maps to reduce potential false positive detections.”
KGSNet: Key-Point-Guided Super-Resolution Network for Pedestrian Detection in the Wild (2020, methodology): “Note that the results of PDOE [38], RepLoss [25] and ALFNet [63] DeepParts [28], F-DNN + SS [20], FasterRCNN + ATT [40], CompACT-Deep [78], SA-FastRCNN [61], RPN + BF [22], MS-CNN [79], GDFL [39], TLL-TFA [41], SDS-RCNN [19], PDOE [38], RepLoss [25], and ALFNet [63].”
Pedestrian Detection: The Elephant In The Room (2020, methodology): “RPN+BF [40] was the first work to use Region Proposal Network (RPN); it used boosted forest for improving pedestrian detection performance.”
Computer Vision for Autonomous Vehicles: Problems, Datasets and State-of-the-Art (2017, methodology): “We refer the interested reader to the survey papers (Benenson et al. (2014); Zhang et al. (2016b)) for an in-depth comparison of pedestrian detection systems on Caltech-USA.”
Relaxed Softmax: Efficient Confidence Auto-Calibration for Safe Pedestrian Detection (2018, background): “Most modern detectors, including variants specialized for the detection of pedestrians [28, 30, 31, 2], are based on the propose & verify paradigm.”
Joint Detection and Identification Feature Learning for Person Search (2016, background): “Recent years, CNNbased pedestrian detectors have also been developed [36, 39].”
In Teacher We Trust: Learning Compressed Models for Pedestrian Detection (2016, background): “We show the effect of these modiﬁcations on student networks of various sizes, and show that they outperform both training from scratch and standard KD.”
CityPersons: A Diverse Dataset for Pedestrian Detection (2017, methodology): “Better results are shown when using the reverse configuration: detections resulted from a convnet are re-scored with decision forests classifier (trained over convnet features) [4, 31].”
Towards Robust CNN-based Object Detection through Augmentation with Synthetic Rain Variations (2019, background): “the basis for many current object detection approaches like YOLO and part of the pedestrian detector in [28].”
Temporal-Context Enhanced Detection of Heavily Occluded Pedestrians (2020, methodology): “Subset Ours Baseline SDS-RCNN [7] RPN+BF[2] A-FRCCN[6] FRCCN[53] Checkerboards[21]”
Multispectral Fusion for Object Detection with Cyclic Fuse-and-Refine Blocks (2020, neutral): “Following [17], the anchor designing strategy is adapted for the pedestrian detection for KAIST dataset: we fix the aspect ratio of each anchor box to 0.”