FCOS: Fully Convolutional One-Stage Object Detection

By Zhi Tian, Chunhua Shen, Hao Chen, Tong He. 2019

6,275 citations recorded by the source corpus.

We propose a fully convolutional one-stage object detector (FCOS) to solve object detection in a per-pixel prediction fashion, analogue to semantic segmentation. Almost all state-of-the-art object detectors such as RetinaNet, SSD, YOLOv3, and Faster R-CNN rely on pre-defined anchor boxes. In contrast, our proposed detector FCOS is anchor box free, as well as proposal free. By eliminating the pre-defined set of anchor boxes, FCOS completely avoids the complicated computation related to anchor boxes such as calculating overlapping during training. More importantly, we also avoid all hyper-parameters related to anchor boxes, which are often very sensitive to the final detection performance. With the only post-processing non-maximum suppression (NMS), FCOS with ResNeXt-64x4d-101 achieves 44.7% in AP with single-model and single-scale testing, surpassing previous one-stage detectors with the advantage of being much simpler. For the first time, we demonstrate a much simpler and flexible detection framework achieving improved detection accuracy. We hope that the proposed FCOS framework can serve as a simple and strong alternative for many other instance-level tasks. Code is available at: https://tinyurl.com/FCOSv1

Subfields: Computer Vision and Pattern Recognition

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

Citations in the corpus

End-to-End Object Detection with Transformers (2020, background): “Modern detectors address this set prediction task in an indirect way, by defining surrogate regression and classification problems on a large set of proposals [37,5], anchors [23], or window centers [53,46].”
Bridging the Gap Between Anchor-Based and Anchor-Free Detection via Adaptive Training Sample Selection (2019, methodology): “FCOS [56] regards all the locations inside the object bounding box as positives with four distances and a novel centerness score to detect objects.”
Pix2seq: A Language Modeling Framework for Object Detection (2021, methodology): “To predict a set of bounding boxes, architectures of modern detectors are speciﬁcally designed to produce a large set of proposals (Girshick, 2015; Ren et al., 2015; Cai & Vasconcelos, 2018), anchors (Lin et al., 2017b), or window centers (Tian et al., 2019; Zhou et al., 2019).”
Grounded Language-Image Pre-training (2021, background): “In one-stage detectors, localization loss Lloc may also contain the centerness loss [54].”
Exploring Plain Vision Transformer Backbones for Object Detection (2022, neutral): “anchor-free [30,14,49], and region-based [20,25,19,46] vs.”
Towards Real-World Prohibited Item Detection: A Large-Scale X-ray Benchmark (2021, background): “Recently, anchor-free approaches have at-tracted wide attention of researchers by using key points to represent the objects, including CornerNet [15], CenterNet [6], and FCOS [34].”
Learning Open-World Object Proposals without Learning to Classify (2021, methodology): “For the box regression, we replace the standard box-delta targets (xyhw) with distances from the location to four sides of the ground-truth box (lrtb) as in [56].”
Occluded Video Instance Segmentation: A Benchmark (2021, methodology): “Based on FCOS [49], SGNet [38] dynamically divides instance into sub-regions and performs segmentation on each region.”
Deformable DETR: Deformable Transformers for End-to-End Object Detection (2020, methodology): “Method Backbone TTA AP AP50 AP75 APS APM APL FCOS (Tian et al., 2019) ResNeXt-101 44.”
Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions (2021, background): “In the era of deep learning, CNNs [34] have become the dominant framework for object detection, which includes single-stage detectors ( e.g. , SSD [43], RetinaNet [39], FCOS [62], GFL [37, 35], Po-larMask [71] and OneNet [55]) and multi-stage detectors (Faster R-CNN [49], Mask R-CNN [21], Cascade R-”
EfficientDet: Scalable and Efficient Object Detection (2019, background): “There have been many previous works aiming to develop more efﬁcient detector architectures, such as one-stage [20, 25, 26, 17] and anchor-free detectors [14, 36, 32], 1 Similar to [9, 31], FLOPS denotes number of multiply-adds.”
Overcoming Catastrophic Forgetting in Incremental Object Detection via Elastic Response Distillation (2022, methodology): “We perform extended experiments to validate the generality of the proposed method on the FCOS detector.”
SID: Incremental Learning for Anchor-Free Object Detection via Selective and Inter-Related Distillation (2020, methodology): “We build our method on top of FCOS (Tian et al., 2019) and CenterNet (Zhou et al., 2019) using their public implementations.”
Generalized Focal Loss: Learning Qualified and Distributed Bounding Boxes for Dense Object Detection (2020, methodology): “Assume that a model estimates probabilities Type FCOS [26] ATSS [31] AP AP50 AP75 APS APM APL AP AP50 AP75 APS APM APL w/o quality branch 37.8 56.2 40.8 21.2 42.1 48.2 38.0 56.5 40.7 20.6 42.1 49.1 centerness-branch [26] 38.5 56.8 41.6 22.4 42.4 49.1 39.2 57.4 42.2 23.0 42.8 51.1 IoU-branch [29, 12]”
MMDetection: Open MMLab Detection Toolbox and Benchmark (2019, methodology): “We benchmark different methods on COCO 2017 val , including SSD [19], RetinaNet [18], Faster RCNN [27], Mask RCNN [13] and Cascade R-CNN [18], Hybrid Task Cascade [4] and FCOS [32].”
Deep High-Resolution Representation Learning for Visual Recognition (2019, methodology): “The results are obtained from MMDetection [17]. tection performance on COCO val . under two anchor-based frameworks: Faster R-CNN [118] and Cascade R-CNN [12], and two recently-developed anchor-free frameworks: FCOS [136] and CenterNet [36].”
End-to-End Semi-Supervised Object Detection with Soft Teacher (2021, methodology): “There are two mainstreams: single-stage object detectors [17, 21, 30] and two-stage object detectors [6, 22, 14, 34, 35].”
Dynamic Head: Unifying Object Detection Heads with Attentions (2021, methodology): “We ﬁnd recent studies [12, 35, 28, 31, 33] only focus on solving one of the aforementioned problems in various ways.”