SCAN: Semantic Context Aware Network for Accurate Small Object Detection

Abstract

Recent deep convolutional neural network-based object detectors have shown promising performance when detecting large objects, but they are still limited in detecting small or partially occluded ones-in part because such objects convey limited information due to the small areas they occupy in images. Consequently, it is difficult for deep neural networks to extract sufficient distinguishing fine-grained features for high-level feature maps, which are crucial for the network to precisely locate small or partially occluded objects. There are two ways to alleviate this problem: the first is to use lower-level but larger feature maps to improve location accuracy and the second is to use context information to increase classification accuracy. In this paper, we combine both methods by first constructing larger and more meaningful feature maps in top-down order and concatenating them and subsequently fusing multilevel contextual information through pyramid pooling to construct context aware features. We propose a unified framework called the Semantic Context Aware Network (SCAN) to enhance object detection accuracy. SCAN is simple to implement and can be trained from end to end. We evaluate the proposed network on the KITTI challenge benchmark and present an improvement of the precision.

YOLO-Fine: One-Stage Detector of Small Objects Under Various Backgrounds in Remote Sensing Images

“…In this review, the authors have mentioned five crucial aspects that are involved in recent small object detection frameworks, including multiscale feature learning, data augmentation, training strategy, context-based learning, and generative network-based detection. They also highlighted some powerful models to detect generic small objects, such as improved Faster R-CNN [35,36], Feature-fused SSD [37], MDSSD (Multi-scale deconvolutional SSD) [38], RefineDet [39], SCAN (Semantic context aware network) [40], etc. In the remote sensing community, the detection of small objects has been mostly tackled by exploiting two-stage object detectors thanks to their capacity to generally provide more accurate detection performance as compared to one-stage detectors.…”

Feature Rescaling and Fusion for Tiny Object Detection

“…To increase the training speed and performance, SNIPPER [19], [41] is proposed in [19] to focus on the context information around ground-truth instance. In [20], a feature map and context fusion method is proposed to set up the relationship between small objects and the environment. Besides the environment, in [21] and [22], the relationship between the objects and the features are modified to improve the detection results.…”

“…The reported results follows Average Precision (AP) metrics, which includes AP all , AP small , AP tiny , AP tiny1 , AP tiny2 and AP tiny3 . The size range (pixels) of the objects are divided into three sub-intervals for TOD dataset: tiny [2,20], small [20,32] and all [2, inf]. And for tiny [2,20], it is partitioned into 3 subintervals: tiny1 [2,8], tiny2 [8,12], tiny3 [12,20].…”

Small Object Detection in Remote Sensing Images Based on Super-Resolution with Auxiliary Generative Adversarial Networks

“…To improve the detection of small objects, e.g., vehicles, ships, and animals in satellite images, conventional state-of-the-art object detectors in computer vision such as Faster R-CNN (Faster Region-based Convolutional Neural Network) [1], SSD (Single Shot Multibox Detector) [2], Feature Pyramid Network [3], Mask R-CNN [4], YOLOv3 (You Only Look Once version 3) [5], EfficientDet [6], or others-see a survey of 20-year object detection in [7]-can be specialized by reducing anchor sizes, using multi-scale feature learning with data augmentation to target these small object sizes. We mention here some recent proposed models to tackle generic small object detection such as the improved Faster R-CNN [8], Feature-fused SSD [9], RefineDet [10], SCAN (Semantic context aware network) [11], etc. For more details about their architectures and other developed models, we refer readers to a recent review on deep learning-based small object detection in the computer vision domain [12].…”