CornerNet-Lite: Efficient Keypoint based Object Detection

By Hei Law, Yun Teng, Olga Russakovsky, Jia Deng. 2019

184 citations recorded by the source corpus.

Keypoint-based methods are a relatively new paradigm in object detection, eliminating the need for anchor boxes and offering a simplified detection framework. Keypoint-based CornerNet achieves state of the art accuracy among single-stage detectors. However, this accuracy comes at high processing cost. In this work, we tackle the problem of efficient keypoint-based object detection and introduce CornerNet-Lite. CornerNet-Lite is a combination of two efficient variants of CornerNet: CornerNet-Saccade, which uses an attention mechanism to eliminate the need for exhaustively processing all pixels of the image, and CornerNet-Squeeze, which introduces a new compact backbone architecture. Together these two variants address the two critical use cases in efficient object detection: improving efficiency without sacrificing accuracy, and improving accuracy at real-time efficiency. CornerNet-Saccade is suitable for offline processing, improving the efficiency of CornerNet by 6.0x and the AP by 1.0% on COCO. CornerNet-Squeeze is suitable for real-time detection, improving both the efficiency and accuracy of the popular real-time detector YOLOv3 (34.4% AP at 30ms for CornerNet-Squeeze compared to 33.0% AP at 39ms for YOLOv3 on COCO). Together these contributions for the first time reveal the potential of keypoint-based detection to be useful for applications requiring processing efficiency.

Subfields: 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 Computer Vision and Pattern Recognition

Citations in the corpus

Bridging the Gap Between Anchor-Based and Anchor-Free Detection via Adaptive Training Sample Selection (2019, background): “2 CornerNet-Lite [27] trainval35k Hourglass-54 43.”
YOLOv4: Optimal Speed and Accuracy of Object Detection (2020, methodology): “The detectors of this sort are CenterNet [13], CornerNet [37, 38], FCOS [78], etc.”
Conditional DETR for Fast Training Convergence (2021, methodology): “Typical methods include YOLOv1 [30], CornerNet [21], ExtremeNet [50], CenterNet [49, 6], FCOS [39] and others [23, 28, 52, 19, 51, 22, 15, 46, 47].”
RelationNet++: Bridging Visual Representations for Object Detection via Transformer Decoder (2020, methodology): “Table 1: Ablation on key selection approaches #keys share AP AP50 AP75 - - 35.6 55.5 39.0 20 7 36.1 54.9 39.6 50 7 37.0 55.8 40.6 20 3 37.7 56.5 41.4 50 3 38.5 57.0 42.3 100 3 38.3 56.9 42.0 200 3 38.2 56.7 41.9 Table 2: Ablation of sub-pixel corner/centers CLS (ct.) REG (cn.) AP AP50 AP75 AP90 - - ”
Scaled-YOLOv4: Scaling Cross Stage Partial Network (2020, background): “Today’s popular one-stage object detectors mainly have two kinds: anchor-based [26, 16] and anchor-free [6, 12, 13, 32].”
A Survey of Deep Learning-Based Object Detection (2019, methodology): “To address that, recently a series of anchor-free methods [51], [61], [62], [107]–[113] are proposed.”
CoDeNet: Efficient Deployment of Input-Adaptive Object Detection on Embedded FPGAs (2021, methodology): “In this work, we use a one-stage anchor-free detector called CenterNet [44] due to its better Pareto efficiency for the speed-accuracy tradeoff compared to the concurrent works [9][17][18][45].”
Rethinking Classification and Localization for Object Detection (2019, methodology): “Point-based methods [21, 22, 47, 7, 48] model an object as keypoints (corner, center, etc), and are built on keypoint estimation networks.”
CSPNet: A New Backbone that can Enhance Learning Capability of CNN (2019, methodology): “Two object detectors of this sort are CenterNet [43] and CornerNet-Lite [14], and they both perform very well in terms of efficiency and efficacy.”
Occluded Prohibited Items Detection: An X-ray Security Inspection Benchmark and De-occlusion Attention Module (2020, methodology): “More recently, [4, 12, 13, 33, 39] solved the problem of traditional methods relying heavily on the anchor box, with the only post-processing non-maximum suppression [19].”