Gradient Harmonized Single-stage Detector

By Buyu Li, Yu Liu, Xiaogang Wang. 2018

597 citations recorded by the source corpus.

Despite the great success of two-stage detectors, single-stage detector is still a more elegant and efficient way, yet suffers from the two well-known disharmonies during training, i.e. the huge difference in quantity between positive and negative examples as well as between easy and hard examples. In this work, we first point out that the essential effect of the two disharmonies can be summarized in term of the gradient. Further, we propose a novel gradient harmonizing mechanism (GHM) to be a hedging for the disharmonies. The philosophy behind GHM can be easily embedded into both classification loss function like cross-entropy (CE) and regression loss function like smooth-L1 (SL1) loss. To this end, two novel loss functions called GHM-C and GHM-R are designed to balancing the gradient flow for anchor classification and bounding box refinement, respectively. Ablation study on MS COCO demonstrates that without laborious hyper-parameter tuning, both GHM-C and GHM-R can bring substantial improvement for single-stage detector. Without any whistles and bells, the proposed model achieves 41.6 mAP on COCO testdev set which surpass the state-of-the-art method, Focal Loss (FL) + SL1, by 0.8. The code1 is released to facilitate future research.

Subfields: Artificial Intelligence, Biomedical Engineering, Media Technology

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 Biomedical Engineering

In Media Technology

Citations in the corpus

EPicker is an exemplar-based continual learning approach for knowledge accumulation in cryoEM particle picking (2022, methodology): “On the one hand, we followed the concept of GHM loss 41 and ignored some hard examples that can be seen as outliers.”
MMDetection: Open MMLab Detection Toolbox and Benchmark (2019, background): “• GHM [16]: a gradient harmonizing mechanism to improve single-stage detectors, proposed in 2019.”
Feature Selective Anchor-Free Module for Single-Shot Object Detection (2019, neutral): “2 GHM800 [18] 39.”
Dynamic Anchor Learning for Arbitrary-Oriented Object Detection (2020, background): “The work (Li, Liu, and Wang 2019) further considers extremely hard samples as outliers, and use gradient harmonizing mechanism to conquer the imbalance problems.”
Learning From Noisy Anchors for One-Stage Object Detection (2019, methodology): “As in [22, 35, 46, 17, 27], input images are resized to have a shorter side of 800 while the longer side is kept less than 1333; we also perform random horizontal image flipping for data augmentation.”
Enhancing Geometric Factors in Model Learning and Inference for Object Detection and Instance Segmentation (2020, background): “In GHM [49], the authors proposed a gradient harmonizing mechanism for bounding box regression loss that rectiﬁes the gradient contributions of samples.”
GS3D: An Efficient 3D Object Detection Framework for Autonomous Driving (2019, background): “Since state-of-the-art 2D detectors [21, 18, 13, 17, 16, 15] can provide reliable 2D bounding boxes for objects,”
Balanced Knowledge Distillation for Long-tailed Learning (2021, background): “Cost-sensitive re-weighting [10, 47, 23] assigns different weights to the classification loss terms corresponding to different classes [7, 23, 5] or different samples [29, 22, 27].”
Imbalance Problems in Object Detection: A Review (2019, background): “[59] showed that this scheme results in a large number of noisy examples.”
IoU-uniform R-CNN: Breaking Through the Limitations of RPN (2019, background): “For example, in order to alleviate the ForgroundBackground imbalance, Focal loss [21], Gradient Harmonizing Mechanism (GHM) [19] solves it by a soft sampling way, which suppresses the gradient originating from easy positives and negatives.”
Prime-Aware Adaptive Distillation (2020, methodology): “First, we note that sample weighting has been widely used in other tasks, such as object detection [20,22] and metric learning [12,37].”
Deep Learning for Generic Object Detection: A Survey (2018, background): “Recently, this problem has also acquired some attention [149, 164, 242].”
Equalization Loss for Long-Tailed Object Recognition (2020, background): “[27, 25] make the neural network to be cost-sensitive by increasing the weight for hard samples and decreasing the weight for easy samples, which can be seen as online versions of hard example mining technique [39].”
PV-RCNN++: Point-Voxel Feature Set Abstraction With Local Vector Representation for 3D Object Detection (2021, unknown): “In this direction, object detectors are further divided into two-stage [2], [26], [27], [28], [29] and one-stage [3], [7], [30], [31] categories.”
IoU-aware Single-stage Object Detector for Accurate Localization (2019, background): “Most of these models can be classified into single-stage object detectors [1, 2, 3, 4, 5, 6] and multi-stage object detectors [7, 8, 9, 10, 11, 12, 13].”
Distance-IoU Loss: Faster and Better Learning for Bounding Box Regression (2019, background): “There are also several loss functions for object detection, e.g., focal loss (Lin et al. 2017), class-balanced loss (Cui et al. 2019), balanced loss for classiﬁcation and bounding box regression (Pang et al. 2019), and gradient ﬂow balancing loss (Li, Liu, and Wang 2019).”
Recent Advances in Deep Learning for Object Detection (2019, background): “[147] adopt a similar idea from focal loss and propose a novel gradient harmonizing mechanism (GHM).”
Learning Imbalanced Datasets with Label-Distribution-Aware Margin Loss (2019, background): “[31] suggests an improved technique which down-weights examples with either very small gradients or large gradients because examples with small gradients are well-classified and those with large gradients tend to be outliers.”