Dynamic R-CNN: Towards High Quality Object Detection via Dynamic Training

By Hongkai Zhang, Hong Chang, Bingpeng Ma, Naiyan Wang, Xilin Chen. European Conference on Computer Vision, 2020

618 citations recorded by the source corpus.

Although two-stage object detectors have continuously advanced the state-of-the-art performance in recent years, the training process itself is far from crystal. In this work, we first point out the inconsistency problem between the fixed network settings and the dynamic training procedure, which greatly affects the performance. For example, the fixed label assignment strategy and regression loss function cannot fit the distribution change of proposals and thus are harmful to training high quality detectors. Consequently, we propose Dynamic R-CNN to adjust the label assignment criteria (IoU threshold) and the shape of regression loss function (parameters of SmoothL1 Loss) automatically based on the statistics of proposals during training. This dynamic design makes better use of the training samples and pushes the detector to fit more high quality samples. Specifically, our method improves upon ResNet-50-FPN baseline with 1.9% AP and 5.5% AP$_{90}$ on the MS COCO dataset with no extra overhead. Codes and models are available at this https URL.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/b9e2f03f866f18f485eaaa2fc9a85c49c5719974

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

Scaled-YOLOv4: Scaling Cross Stage Partial Network (2020, background): “Through analysis of state-of-the-art object detectors [1, 3, 5, 22, 31, 36, 40], we found that CSPDarknet53, which is the backbone of YOLOv4 [1], matches almost all optimal architecture features obtained by network architecture search technique.”
AFPN: Asymptotic Feature Pyramid Network for Object Detection (2023, methodology): “We evaluated Faster R-CNN [8] and Dynamic R-CNN [7] in our study.”
QueryDet: Cascaded Sparse Query for Accelerating High-Resolution Small Object Detection (2021, neutral): “2 Dynamic R-CNN [65] ECCV 2020 ResNet-101 42.”
Sparse R-CNN: End-to-End Object Detection with Learnable Proposals (2020, background): “2) The many-to-one label assignment problem [2, 46, 47] in training makes the network sensitive to heuristic assign rules.”
A Review on Deep Learning in UAV Remote Sensing (2021, neutral): “Other state-of-the-art representatives of such algorithms are Cascade-RCNN [32], Trident-Net [185], Grid-RCNN [71], Dynamic-RCNN [52], De-tectoRS [44].”
Localization Distillation for Object Detection (2022, methodology): “R-CNN series [4], [47], [56], [79] adopt multiple regression stages to refine the detection results, while YOLO series [3], [53], [54], [55], SSD series [14], [42], [89], and FCOS series [38], [62] adopt one-stage regression.”
Dynamic Anchor Learning for Arbitrary-Oriented Object Detection (2020, methodology): “ATSS (Zhang et al. 2020b) has achieved great success in object detection using horizontal box.”
Focal and Efficient IOU Loss for Accurate Bounding Box Regression (2021, methodology): “Current state-of-the-art object detectors (e.g., Cascade R-CNN [2], Mask R-CNN [9] and Dynamic R-CNN [26]) rely on a bounding box regression (BBR) module to localize objects.”
MPDIoU: A Loss for Efficient and Accurate Bounding Box Regression (2023, methodology): “GIoU loss has been applied to train state-of-the-art object detectors, such as YOLO v3 and Faster R-CNN, and achieves better performance than MSE loss and IoU loss.”