FCOS: A Simple and Strong Anchor-free Object Detector

Tian, Zhi; Shen, Chunhua; Chen, Hao; He, Tong

doi:10.1109/tpami.2020.3032166

Cited by 275 publications

(212 citation statements)

References 50 publications

Supporting

Mentioning

206

Contrasting

Order By: Relevance

“…In addition, CornerNet [20] and Cen-terNet [21], [22] get rid of anchor boxes with the help of key points and heatmap prediction, thus performing more flexible object detection. FCOS [8] can directly predict the distance to four sides of the object box in each pixel prediction, and simultaneously categorize each pixel. At present, FCOS has been widely applied to various fields to solve problems in object detection because of its simplicity and efficiency.…”

Section: Related Work a Image Object Detectionmentioning

confidence: 99%

“…Image object detection algorithm will analyze a given input image and output the category and accurate localization of each object contained in the image. In recent years, with rapid development of convolutional neural network, the object detection algorithms [1]- [8] based on deep convolutional neural network have made a great progress. At present, as the basic task of computer vision, object detection algorithm has been widely applied to the industry and our life.…”

Section: Introductionmentioning

confidence: 99%

“…To validate the effectiveness of AdaFPN proposed herein, two classical object detection algorithms, i.e., Faster R-CNN and FCOS [8], were employed in this paper as experimental benchmarks. Faster R-CNN, as a classical twostage algorithm, predicted the proposals containing objects at the first stage, and extracted region features corresponding to each proposal by RoI Pooling [1] for classification and regression at the second stage.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Adaptive Feature Pyramid Networks for Object Detection

Wang

Zhong

2021

IEEE Access

View full text Add to dashboard Cite

In general object detection, scale variation is always a big challenge. At present, feature pyramid networks are employed in numerous methods to alleviate the problems caused by large scale range of objects in object detection, which makes use of multi-level features extracted from the backbone for top-down upsampling and fusion to acquire a set of multi-scale depth image features. However, the feature pyramid network proposed by Lin et al. adopts a simple fusion method, which fails to consider the fusion feature context, and therefore, it is difficult to acquire good features. In addition, the fusion of multi-scale features directly by traditional upsampling is prone to feature misalignment and loss of details. In this paper, an adaptive feature pyramid network is proposed based on the feature pyramid network to alleviate the foregoing potential problems, which includes two major designs, i.e., adaptive feature upsampling and adaptive feature fusion. The adaptive feature upsampling aims to predict a group of sampling points of each pixel through some models, and constitute feature representation of the pixel by feature combination of sampling points, while adaptive feature fusion is to construct pixel-level fusion weights between fusion features through attention mechanism. The experimental results verified the effectiveness of the method proposed in this paper. On the public object detection dataset MS-COCO test-dev, Faster R-CNN model achieved performance improvement of 1.2 AP by virtue of the adaptive feature pyramid network, and FCOS model could achieve performance improvement of 1.0 AP. What's more, the experiments also validated that the adaptive feature pyramid network proposed herein was more accurate for object localization.

show abstract

Section: Related Work a Image Object Detectionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Adaptive Feature Pyramid Networks for Object Detection

Wang

Zhong

2021

IEEE Access

View full text Add to dashboard Cite

show abstract

“…However, the result of our model is more accurate than the Mixed YOLOv3-LITE model even when trained with the input size of 416 and still maintain the real-time speed. UWS-YOLO was also compared with Fully Convolutional One-Stage (FCOS) object detector (Tian et al 2020), CenterNet (Duan et al 2019), CornerNet (Law and Deng 2020), RefineNet (Zhang et al 2018) and RetinaNet (Lin et al 2020) detectors in Table 6 using VisDrone-DET 2019 (Du et al 2019). In terms of FCOS being trained with VisDrone-DET 2019, the accuracy of 28.8% was obtained using ResNet50 with resolution 2666 which is less than our approach (29.8%) with a resolution of 416.…”

Section: Results With Visdrone2019mentioning

confidence: 93%

Efficient Real-Time Human Detection Using Unmanned Aerial Vehicles Optical Imagery

Golkarnarenji

Martinez‐Alpiste

Wang

et al. 2021

International Journal of Remote Sensing

View full text Add to dashboard Cite

Unmanned Aerial Vehicles (UAVs) are promising technologies within many different application scenarios including human detection in search and rescue and surveillance use cases, which have received considerable attention worldwide. However, adverse conditions, such as varying altitude, overhead camera placement, changing illumination and moving platform, impose challenges for highperformance yet cost-efficient human detection. To overcome these challenges, we propose a novel combination of dilated convolutions with Path Aggregation Network (PAN) as a new deep neural network-based human detection algorithm in real time. Furthermore, we establish a comprehensive human detection dataset with varying backgrounds, illuminations, and contrast and train the proposed machine-learning model on the collected dataset. Our approach achieves both high precision (88.0% mean Average Precision (mAP)) and real time (67.0 Frames Per Second (FPS)) on a commercial off-the-shelf PC platform. In terms of accuracy, the result is comparable to the standard You Only Look Once v3 (YOLOv3). However, the speed is twice as that of the standard YOLOv3. YOLOv4 is slightly more accurate (89.8%) than our approach. However, it is slower (38.0 versus 67.0 FPS) and has more Billion Floating-Point Operations (BFLOPS). The proposed algorithm has also trained with the VisDrone2019 dataset and compared with seven studies using this dataset. The results have further validated the effectiveness of the proposed approach. Moreover, the algorithm has been evaluated on an embedded system (Jetson AGX Xavier), which demonstrates the usefulness of this method on power-constrained devices. The proposed algorithm is fast, memory efficient, and computationally less expensive to achieve high detection performance. It is expected to contribute significantly to the wider use of UAV applications including search and rescue missions to locate missing people, and surveillance particularly for applications running on resource-constrained platforms, like smartphones or tablets. This proposed system is now being used in aerial drone system of Police of Scotland to help them locate and find missing and vulnerable people. The results of the project were broadcasted by BBC Scotland.

show abstract

“…The networks currently widely used are anchor-based and anchorfree. Typical anchor-free neworks, whether keypoint-based (CornerNet [17] and CoenterNet [18]) or pixel-wise prediction (FCOS [19]), are essentially dense prediction methods. Large and flexible solution space makes anchorfree model obtain high accuracy, but it is prone to semantic confusion and unstable detection results for densely distributed objects.…”

Section: Anchor Strategymentioning

confidence: 99%

Vehicle Detection in Congested Traffic Based on Simplified Weighted Dual-Path Feature Pyramid Network With Guided Anchoring

et al. 2021

View full text Add to dashboard Cite

In modern life, traffic congestion is widespread in large and medium-sized cities in various countries. Multi-scale vehicle targets are densely distributed and occluded from each other in the images of crowded scenes. Vehicle detection in such scenarios is of great significance to urban traffic control, safety management and criminal investigation, but also has great challenges. Facing the special application in congested traffic, we propose Simplified Weighted Dual-path Network with Guided Anchoring framework to realize real-time vehicle detection. Firstly, a simplified weighted Dual-path Feature Pyramid Network (SWD-FPN) is used to improve the robustness of the model for multi-scale and partially occluded objects. Secondly, in order to improve the detection capability for vehicles with wide range of scale changes, the Guided Anchoring (GA) is applied to generate anchors of corresponding positions and scales according to the feature maps. Finally, for the challenge of vehicles intensive distribution, DIoU-soft NMS postprocessing mechanism is introduced to reduce the missing alarm. Considering the class imbalance of vehicle detection in real-time traffic scenarios and the above improvements, multi-task loss is proposed for training. Ablation experiments are performed on UA-DETRAC dataset to further analyze the effect of different strategies on performance improvement. In addition, comparisons experiments on UA-DETRAC dataset and handcrafted Vehicles of Traffic (VOF) dataset are conducted to demonstrate the superiority of the proposed method over other state-of-the-art methods.

show abstract

FCOS: A Simple and Strong Anchor-free Object Detector

Cited by 275 publications

References 50 publications

Adaptive Feature Pyramid Networks for Object Detection

Adaptive Feature Pyramid Networks for Object Detection

Efficient Real-Time Human Detection Using Unmanned Aerial Vehicles Optical Imagery

Vehicle Detection in Congested Traffic Based on Simplified Weighted Dual-Path Feature Pyramid Network With Guided Anchoring

Contact Info

Product

Resources

About