DetNet: A Backbone network for Object Detection

By Zeming Li, Chao Peng, Gang Yu, Xiangyu Zhang, Yangdong Deng. 2018

280 citations recorded by the source corpus.

Recent CNN based object detectors, no matter one-stage methods like YOLO, SSD, and RetinaNe or two-stage detectors like Faster R-CNN, R-FCN and FPN are usually trying to directly finetune from ImageNet pre-trained models designed for image classification. There has been little work discussing on the backbone feature extractor specifically designed for the object detection. More importantly, there are several differences between the tasks of image classification and object detection. 1. Recent object detectors like FPN and RetinaNet usually involve extra stages against the task of image classification to handle the objects with various scales. 2. Object detection not only needs to recognize the category of the object instances but also spatially locate the position. Large downsampling factor brings large valid receptive field, which is good for image classification but compromises the object location ability. Due to the gap between the image classification and object detection, we propose DetNet in this paper, which is a novel backbone network specifically designed for object detection. Moreover, DetNet includes the extra stages against traditional backbone network for image classification, while maintains high spatial resolution in deeper layers. Without any bells and whistles, state-of-the-art results have been obtained for both object detection and instance segmentation on the MSCOCO benchmark based on our DetNet~(4.8G FLOPs) backbone. The code will be released for the reproduction.

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

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

In Model Architectures

Citations in the corpus

Rethinking ImageNet Pre-Training (2018, background): “DetNet [25] and CornerNet [22] also present results when training detectors from scratch.”
Group Normalization (2018, background): “Here the BN results are from [34], and BN is synced across GPUs [43] and is not frozen.”
Bridging the Gap Between Anchor-Based and Anchor-Free Detection via Adaptive Training Sample Selection (2019, neutral): “6 DetNet [31] trainval35k DetNet-59 40.”
NAS-FPN: Learning Scalable Feature Pyramid Architecture for Object Detection (2019, background): “In the past few years, great progress has been made on designing the model architecture of deep convolutional networks (ConvNets) for image classification [12, 15, 35] and object detection [21, 22].”
Feature Selective Anchor-Free Module for Single-Shot Object Detection (2019, background): “DetNet [19] designed a novel backbone network to maintain high spatial resolution in upper pyramid levels.”
MetaAnchor: Learning to Detect Objects with Customized Anchors (2018, background): “Theoretically MetaAnchor could work with most of the existing anchor-based object detection frameworks [32, 25, 30, 31, 23, 11, 22, 19, 20, 2].”
Deep Learning for Generic Object Detection: A Survey (2018, methodology): “rmance at minimal additional computational cost. on CNN architectures remain active, and a numer of backbone networks are still emerging such as Dilated Residual Networks [230], Xception [35], DetNet [127], and Dual Path Networks (DPN) [31]. The training of a CNN requires a large labelled dataset wi”
DSFD: Dual Shot Face Detector (2018, result): “Therefore, in our DSFD framework, better performance on classiﬁcation are not necessary for better performance on detection, which is consistent to the conclusion claimed in [11, 16].”
Objects365: A Large-Scale, High-Quality Dataset for Object Detection (2019, unknown): “De-formable ConvNet [4, 39], Soft Sampling [34], SNIP [32], SNIPER [33], and Cascade R-CNN [1], DetNet [21] are introduced to further improve the performance.”
YOLO-FaceV2: A Scale and Occlusion Aware Face Detector (2022, background): “A natural idea is to fuse deep semantic information with shallow features, such as [20, 41, 42].”
Wav2KWS: Transfer Learning From Speech Representations for Keyword Spotting (2021, methodology): “To achieve a high KWS performance, we adopt a suitable encoder for learning transferred speech representations from a backbone [24], as shown in Fig.”
Recent Advances in Deep Learning for Object Detection (2019, background): “However, there still exist conﬂicts between classiﬁcation and detection tasks [78], and thus it’s not an optimal solution to do so.”
Object Detection in 20 Years: A Survey (2019, methodology): “Dilated convolution has now been widely used in object detection [21, 56, 254, 255], and proves to be effective for improved accuracy without any additional parameters and computational cost [56].”
GCNet: Non-Local Networks Meet Squeeze-Excitation Networks and Beyond (2019, background): “As convolution networks have recently achieved great success in large-scale visual recognition tasks, a number of attempts have been made to improve the original architecture in a bid to achieve better accuracy [19, 26, 27, 11, 37, 16, 34, 15, 43, 14, 40, 12, 5, 42, 20, 2, 24, 31, 35, 7].”
VisDrone-DET2019: The Vision Meets Drone Object Detection in Image Challenge Results (2019, methodology): “Except Cascade R-CNN [2], The Vis-Drone team also gives the results of another 6 baseline methods, i.e. , CornerNet [18], Light-RCNN [23], Det-Net59 [24], ReﬁneDet [45], RetinaNet [26] and FPN [25].”
SP-NAS: Serial-to-Parallel Backbone Search for Object Detection (2020, methodology): “…detection-speciﬁc backbone networks [38, 42, 18] follows a serial structure and try to leverage and preserve the spatial information by adding dilation and a higher spatial resolution in the later stage (DetNet [18]), or preserving and reﬁning features of different resolutions (FishNet [38]).”
Computation Reallocation for Object Detection (2019, methodology): “To reallocate the computation across spatial position, we utilize dilated convolution Li et al. (2019), Li et al. (2018).”
A Survey of Deep Learning-Based Object Detection (2019, background): “[16] propose a newly designed object detection architecture to specially focus on detection task which maintains high spatial resolution in deeper layers and”
ScratchDet: Exploring to Train Single-Shot Object Detectors from Scratch (2018, background): “Recent developments of two-stage approach focus on redesigning architecture diagram (Li et al. 2018), convolution form (Dai et al.”