Revisiting AP Loss for Dense Object Detection: Adaptive Ranking Pair Selection

  title={Revisiting AP Loss for Dense Object Detection: Adaptive Ranking Pair Selection},
  author={Dongli Xu and Jinhong Deng and Wen Li},
  journal={2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  • Dongli XuJinhong DengWen Li
  • Published 1 June 2022
  • Computer Science
  • 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)
Average precision (AP) loss has recently shown promising performance on the dense object detection task. However, a deep understanding of how AP loss affects the detector from a pairwise ranking perspective has not yet been developed. In this work, we revisit the average precision (AP) loss and reveal that the crucial element is that of selecting the ranking pairs between positive and negative samples. Based on this observation, we propose two strategies to improve the AP loss. The first of… 

Figures and Tables from this paper



RankDetNet: Delving into Ranking Constraints for Object Detection

This work proposes a ranking-based optimization algorithm for harmoniously learning to rank and localize proposals in lieu of the classification task, and comprehensively investigates three types of ranking constraints, i.e., global ranking, class-specific ranking and IoU-guided ranking losses.

A Ranking-based, Balanced Loss Function Unifying Classification and Localisation in Object Detection

A unified, bounded, balanced and ranking-based loss function for both classification and localisation tasks in object detection that naturally enforces high-quality localisation for high-precision classification.

Rank & Sort Loss for Object Detection and Instance Segmentation

To tackle the non-differentiable nature of ranking and sorting, the incorporation of error-driven update with back-propagation as Identity Update is reformulate, which enables us to model the authors' novel sorting error among positives.

Towards Accurate One-Stage Object Detection With AP-Loss

A novel optimization algorithm is developed, which seamlessly combines the error-driven update scheme in perceptron learning and backpropagation algorithm in deep networks, and good convergence property of the proposed algorithm is verified theoretically and empirically.

Learning to Rank Proposals for Object Detection

A novel Learning-to-Rank (LTR) model is proposed to produce the suppression rank via a learning procedure, thus facilitating the candidate generation and lifting the detection performance.

Focal Loss for Dense Object Detection

This paper proposes to address the extreme foreground-background class imbalance encountered during training of dense detectors by reshaping the standard cross entropy loss such that it down-weights the loss assigned to well-classified examples, and develops a novel Focal Loss, which focuses training on a sparse set of hard examples and prevents the vast number of easy negatives from overwhelming the detector during training.

Probabilistic Anchor Assignment with IoU Prediction for Object Detection

A novel anchor assignment strategy that adaptively separates anchors into positive and negative samples for a ground truth bounding box according to the model's learning status such that it is able to reason about the separation in a probabilistic manner is proposed.

Association Loss for Visual Object Detection

A novel association loss is proposed, namely, the proxy squared error (PSE) loss, to entangle the two sub-nets, thus use the dependency between the classification and localization scores obtained from these twoSub-nets to improve the detection performance.

Learning From Noisy Anchors for One-Stage Object Detection

This paper proposes to mitigate noise incurred by imperfect label assignment such that the contributions of anchors are dynamically determined by a carefully constructed cleanliness score associated with each anchor, and conducts extensive experiments on COCO.

Generalized Focal Loss: Learning Qualified and Distributed Bounding Boxes for Dense Object Detection

Improved representations of quality estimation into the class prediction vector to form a joint representation of localization quality and classification, and use a vector to represent arbitrary distribution of box locations are designed.