Learning Complexity-Aware Cascades for Deep Pedestrian Detection

By Zhaowei Cai, Mohammad Javad Saberian, Nuno Vasconcelos. IEEE International Conference on Computer Vision, 2015

347 citations recorded by the source corpus.

The design of complexity-aware cascaded detectors, combining features of very different complexities, is considered. A new cascade design procedure is introduced, by formulating cascade learning as the Lagrangian optimization of a risk that accounts for both accuracy and complexity. A boosting algorithm, denoted as complexity aware cascade training (CompACT), is then derived to solve this optimization. CompACT cascades are shown to seek an optimal trade-off between accuracy and complexity by pushing features of higher complexity to the later cascade stages, where only a few difficult candidate patches remain to be classified. This enables the use of features of vastly different complexities in a single detector. In result, the feature pool can be expanded to features previously impractical for cascade design, such as the responses of a deep convolutional neural network (CNN). This is demonstrated through the design of a pedestrian detector with a pool of features whose complexities span orders of magnitude. The resulting cascade generalizes the combination of a CNN with an object proposal mechanism: rather than a pre-processing stage, CompACT cascades seamlessly integrate CNNs in their stages. This enables state of the art performance on the Caltech and KITTI datasets, at fairly fast speeds.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/c3bd3b9782dc504ee4f2b8a12bd9c562a5c0d7ad

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

Unsupervised Hard Example Mining from Videos for Improved Object Detection (2018, background): “Many of these object detectors have been re-purposed for other tasks such as face detection [39,29,60,15], [31,63,62,25,57,23,66] and pedestrian detection [64,14,6,7,22,30,65], achieving impressive results [24,61,12].”
A Unified Multi-scale Deep Convolutional Neural Network for Fast Object Detection (2016, methodology): “Two main streams of research have been pursued to improve its speed: fast feature extraction [1,2] and cascade learning [14,17,15].”
UA-DETRAC: A new benchmark and protocol for multi-object detection and tracking (2015, methodology): “, DPM [22], ACF [18], R-CNN [26], and CompACT [11], with the largest F-score, the average execution speeds on 40 sequences in UA-DETRAC-test set are presented in Table 7.”
3D Object Proposals Using Stereo Imagery for Accurate Object Class Detection (2016, background): “[45] designs a complexity-aware cascade pedestrian detector with convolutional features.”
Focus: Querying Large Video Datasets with Low Latency and Low Cost (2018, neutral): “, [29, 43, 56, 83]) to reduce object detection latency.”
Occlusion-aware R-CNN: Detecting Pedestrians in a Crowd (2018, background): “In [6], a complexity-aware cascaded detector is proposed for an optimal trade-off between accuracy and speed.”
Temporal-Context Enhanced Detection of Heavily Occluded Pedestrians (2020, methodology): “To this end, given a pair of bt t and b ki i , the PRM module first applies a segmentation subnetwork Nseg to x kt t to predict the visible mask vt = Nseg(x kt t ) for the current pedestrian candidate, where vt ∈ [0, 1].”
Object Detection With Deep Learning: A Review (2018, methodology): “Prior to the recent progress in DCNN based methods [195], [196], some researchers combined boosted decision forests with hand-crafted features to obtain pedestrian detectors [197]–[199].”
Joint Detection and Identification Feature Learning for Person Search (2016, background): “[2] proposed a complexityaware boosting algorithm for learning CNN detector cascades.”
High-Speed tracking-by-detection without using image information (2017, methodology): “Although CompACT shows a much better average precision of the PR curve (see [17] for further details) than the other reference detections, better PR-MOTA values can be achieved with ACF and R-CNN.”
Learning Efficient Single-Stage Pedestrian Detectors by Asymptotic Localization Fitting (2018, neutral): “4 CompACT-Deep [5] Tesla K40 GPU x1 0.”
IDK Cascades: Fast Deep Learning by Learning not to Overthink (2017, background): “[7] also examined cascades for pedestrian detection, proposing a complexity aware term to regularize the cascade objective.”
Reducto: On-Camera Filtering for Resource-Efficient Real-Time Video Analytics (2020, background): “Key to the success of such applications has been recent advances in computer vision, particularly neural network (NN)-based techniques for highly accurate object detection and recognition [22, 45, 47, 49, 65].”
DETRAC: A New Benchmark and Protocol for Multi-Object Tracking (2015, methodology): “We evaluate the following four stateof-the-art object detection algorithms including DPM [18], ACF [13], R-CNN [21], and CompACT [8].”
Scale-Aware Fast R-CNN for Pedestrian Detection (2015, methodology): “We compare the result of SAF R-CNN with all the existing methods that achieved best performance on the Caltech testing set, including VJ [37], HOG [5], LDCF [22], Katamari [2], SpatialPool-ing+ [25], TA-CNN [35], Checkerboards [44], and CompACT-Deep [46].”
Not All Pixels Are Equal: Difficulty-Aware Semantic Segmentation via Deep Layer Cascade (2017, background): “Network cascades [2, 18, 26, 33, 24] have been studied to improve the performance in classi-ﬁcation [26], detection [18], and pose estimation [33].”
Perceptual Generative Adversarial Networks for Small Object Detection (2017, methodology): “We use dense sampling of the training data (every 4th frame) as adopted in [44, 27].”
Bi-box Regression for Pedestrian Detection and Occlusion Estimation (2018, background): “Many efforts have been made to improve its performance in recent years [3, 8, 17, 6, 33, 40, 5, 39, 41, 37, 34, 4].”
Monocular 3D Object Detection for Autonomous Driving (2016, neutral): “78 CompACT-Deep [7] - 70.”
Pedestrian Detection: The Elephant In The Room (2020, background): “The same trend is also true for the pedestrian detection [2,17,7].”