Soft Anchor-Point Object Detection

By Chenchen Zhu, Fangyi Chen, Zhiqiang Shen, Marios Savvides. 2019

162 citations recorded by the source corpus.

Recently, anchor-free detection methods have been through great progress. The major two families, anchor-point detection and key-point detection, are at opposite edges of the speed-accuracy trade-off, with anchor-point detectors having the speed advantage. In this work, we boost the performance of the anchor-point detector over the key-point counterparts while maintaining the speed advantage. To achieve this, we formulate the detection problem from the anchor point's perspective and identify ineffective training as the main problem. Our key insight is that anchor points should be optimized jointly as a group both within and across feature pyramid levels. We propose a simple yet effective training strategy with soft-weighted anchor points and soft-selected pyramid levels to address the false attention issue within each pyramid level and the feature selection issue across all the pyramid levels, respectively. To evaluate the effectiveness, we train a single-stage anchor-free detector called Soft Anchor-Point Detector (SAPD). Experiments show that our concise SAPD pushes the envelope of speed/accuracy trade-off to a new level, outperforming recent state-of-the-art anchor-free and anchor-based detectors. Without bells and whistles, our best model can achieve a single-model single-scale AP of 47.4% on COCO.

Subfields: Computer Vision and Pattern Recognition, Industrial and Manufacturing Engineering

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

In Industrial and Manufacturing Engineering

Citations in the corpus

Generalized Focal Loss: Learning Qualified and Distributed Bounding Boxes for Dense Object Detection (2020, methodology): “GFL with ResNet-101 [9] achieves 45.0% AP at 14.6 FPS, which is superior than all the existing detectors with the same backbone, including SAPD [33] (43.5”
YOLOv4: Optimal Speed and Accuracy of Object Detection (2020, background): “Soft Anchor-Point Object Detection [101] SAPD ResNet-50 - 14.”
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].”
Generalized Focal Loss V2: Learning Reliable Localization Quality Estimation for Dense Object Detection (2020, methodology): “FPS values with ∗ are from [44] or their official repositories [27], while others are measured on the same machine with a single GeForce RTX 2080Ti GPU under the same mmdetection [3] framework, using a batch size of 1 whenever possible.”
TOOD: Task-aligned One-stage Object Detection (2021, methodology): “For example, FSAF [35] selects meaningful samples from feature pyramids based on the computed loss, and similarly, SAPD [34] provides a soft-selection version of FSAF by designing a meta-selection network.”
OTA: Optimal Transport Assignment for Object Detection (2021, unknown): “9 SAPD [52] 180k ResNeXt-64x4d-101-DCN 47.”
Corner Proposal Network for Anchor-free, Two-stage Object Detection (2020, methodology): “Point-grouping detectors, including CornerNet [19], CenterNet [8], ExtremeNet [49], etc, group more than one keypoints into an object, while the pointvector detectors such as FCOS [39], CenterNet [48], FoveaBox [17], SAPD [50], etc.”
PP-YOLO: An Effective and Efficient Implementation of Object Detector (2020, methodology): “Anchor-point based detectors perform object bounding box regression based on anchor points instead of anchor boxes, including FSAF [49], FCOS[36], FoveaBox[17], SAPD[48].”
VarifocalNet: An IoU-aware Dense Object Detector (2020, methodology): “The popular methods include DenseBox [24], FASF [25], Fove-aBox [13], FCOS [9], ATSS [26] and SPAD [27].”
Probabilistic two-stage detection (2021, methodology): “While one-stage detectors have achieved competitive performance (Zhang et al., 2020b; Kim & Lee, 2020; Zhang et al., 2019; Li et al., 2020b; Zhu et al., 2020a), they usually rely on heavier separate classiﬁcation and regression branches than two-stage models.”