This website requires JavaScript.

Towards Large-Scale Small Object Detection: Survey and Benchmarks

Gong ChengXiang YuanXiwen YaoKebing YanQinghua ZengJunwei Han
Jul 2022
摘要
With the rise of deep convolutional neural networks, object detection hasachieved prominent advances in past years. However, such prosperity could notcamouflage the unsatisfactory situation of Small Object Detection (SOD), one ofthe notoriously challenging tasks in computer vision, owing to the poor visualappearance and noisy representation caused by the intrinsic structure of smalltargets. In addition, large-scale dataset for benchmarking small objectdetection methods remains a bottleneck. In this paper, we first conduct athorough review of small object detection. Then, to catalyze the development ofSOD, we construct two large-scale Small Object Detection dAtasets (SODA),SODA-D and SODA-A, which focus on the Driving and Aerial scenariosrespectively. SODA-D includes 24704 high-quality traffic images and 277596instances of 9 categories. For SODA-A, we harvest 2510 high-resolution aerialimages and annotate 800203 instances over 9 classes. The proposed datasets, aswe know, are the first-ever attempt to large-scale benchmarks with a vastcollection of exhaustively annotated instances tailored for multi-category SOD.Finally, we evaluate the performance of mainstream methods on SODA. We expectthe released benchmarks could facilitate the development of SOD and spawn morebreakthroughs in this field. Datasets and codes will be available soon at:\url{https://shaunyuan22.github.io/SODA}.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答