FCOS: Fully Convolutional One-Stage Object Detection

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

doi:10.48550/arxiv.1904.01355

Cited by 132 publications

(222 citation statements)

References 26 publications

Supporting

Mentioning

207

Contrasting

Unclassified

Order By: Relevance

“…In Tab. 7, we show that our proposed method, in addition to a single-stage anchor-based detector (RetinaNet [13]), can work for other detection models as well: two-stage Faster R-CNN [19], anchor-free FCOS [22], and a single-stage detector with an alternative anchor assignment scheme PAA [7].…”

Section: Experiments and Discussionmentioning

confidence: 99%

NOD: Taking a Closer Look at Detection under Extreme Low-Light Conditions with Night Object Detection Dataset

Morawski¹,

Chen²,

Lin³

et al. 2021

Preprint

View full text Add to dashboard Cite

Recent work indicates that, besides being a challenge in producing perceptually pleasing images, low light proves more difficult for machine cognition than previously thought. In our work, we take a closer look at object detection in low light. First, to support the development and evaluation of new methods in this domain, we present a high-quality large-scale Night Object Detection (NOD) dataset showing dynamic scenes captured on the streets at night. Next, we directly link the lighting conditions to perceptual difficulty and identify what makes low light problematic for machine cognition. Accordingly, we provide instance-level annotation for a subset of the dataset for an indepth evaluation of future methods. We also present an analysis of the baseline model performance to highlight opportunities for future research and show that low light is a non-trivial problem that requires special attention from the researchers. Further, to address the issues caused by low light, we propose to incorporate an image enhancement module into the object detection framework and two novel data augmentation techniques. Our image enhancement module is trained under the guidance of the object detector to learn image representation optimal for machine cognition rather than for the human visual system. Finally, experimental results confirm that the proposed method shows consistent improvement of the performance on low-light datasets.

show abstract

Section: Experiments and Discussionmentioning

confidence: 99%

NOD: Taking a Closer Look at Detection under Extreme Low-Light Conditions with Night Object Detection Dataset

Morawski¹,

Chen²,

Lin³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Label Assignment in Object Detection. The task of selecting which anchors are to be assigned as positive or negative samples has recently been recognized as a crucial factor that greatly affects the detector's performance [8,14,6,9,13]. CenterNet [21] and FoveaBox [7], they both use centersampling strategy to select the positive samples.…”

Section: Related Workmentioning

confidence: 99%

“…As mentioned in 1, previous sampling strategies [14,15,19,6] suffer from these limitations above. Thus, the key idea is to introduce a robust instance-wise distribution, which is prediction-aware and is adapted to the semantic pattern of an object.…”

Section: Formulation Of Quality Distribution Encodermentioning

confidence: 99%

“…2. In our experiments, we adopt a simple anchor-free object detector FCOS [14] as our baseline. The network consists of a backbone, a feature pyramid, and one detection head per pyramid level, in a fully convolutional network.…”

Section: Iqdetmentioning

confidence: 99%

“…(2) From Sample-wise to Instance-wise: Assigned by the Sample-wise rule like RetinaNet [8], a predefined anchor is assigned as a positive sample when its IoU is higher than the threshold. Meanwhile, based on Instance-wise distribution, [14,15] propose a center-based distribution according to the ground-truth(GT)'s location, where the samples located inside the central region are assigned as positive samples. These two tendencies achieve better performance and gradually become the de-facto standard in object detection.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

IQDet: Instance-wise Quality Distribution Sampling for Object Detection

Ma¹,

Liu²,

Li³

et al. 2021

Preprint

View full text Add to dashboard Cite

We propose a dense object detector with an instancewise sampling strategy, named IQDet. Instead of using human prior sampling strategies, we first extract the regional feature of each ground-truth to estimate the instance-wise quality distribution. According to a mixture model in spatial dimensions, the distribution is more noise-robust and adapted to the semantic pattern of each instance. Based on the distribution, we propose a quality sampling strategy, which automatically selects training samples in a probabilistic manner and trains with more high-quality samples. Extensive experiments on MS COCO show that our method steadily improves baseline by nearly 2.4 AP without bells and whistles. Moreover, our best model achieves 51.6 AP, outperforming all existing state-of-the-art one-stage detectors and it is completely cost-free in inference time.

show abstract

An image dataset of fusulinid foraminifera generated with the aid of deep learning

Huang

Shi

Chen

et al. 2023

Geoscience Data Journal

View full text Add to dashboard Cite

Fusulinid foraminifera are among the most common microfossils of the Late Palaeozoic and act as key fossils for stratigraphic correlation, paleogeographic and paleoenvironmental indication, and evolutionary studies of marine life. Accurate and efficient identification forms the basis of such research involving fusulinids but is limited by the lack of digitized image datasets. This article presents the first large image dataset of fusulinids containing 2,400 images of individual samples subjected to 16 genera of all six fusulinid families and labelled to species level. These images were collected from the literature and our unpublished samples through an automatic segmentation procedure implementing BlendMask, a deep learning model. The dataset shows promise for the efficient accumulation of fossil images through automated procedures and will facilitate taxonomists in future morphologic and systematic studies.

show abstract

FCOS: Fully Convolutional One-Stage Object Detection

Cited by 132 publications

References 26 publications

NOD: Taking a Closer Look at Detection under Extreme Low-Light Conditions with Night Object Detection Dataset

NOD: Taking a Closer Look at Detection under Extreme Low-Light Conditions with Night Object Detection Dataset

IQDet: Instance-wise Quality Distribution Sampling for Object Detection

An image dataset of fusulinid foraminifera generated with the aid of deep learning

Contact Info

Product

Resources

About