Good Practice on Deep Scene Classification: from Local Supervision to Knowledge Guided Disambiguation
文献类型：会议论文
| 作者 | Yu Qiao; Limin Wang; Sheng Guo; Zhe Wang; Weilin Huang; Yali Wang |
| 出版日期 | 2017 |
| 会议地点 | 美国 |
| 英文摘要 | Recent years witness remarkable progresses of scene classification task, mainly with deep convolutional neural network trained on large scale datasets like Place [7]. This short paper summarize our recent works toward improving the performance of large scale scene classification with deep networks [4, 5, 1]. These works include: 1) encoding locally-supervised convolutional features for scene representation, 2) weakly training patch-level CNNs to extract local discriminative descriptors, and 3) exploiting knowledge from extra networks to release label ambiguity problem of scene categories. We will describe the key insights, approaches and results of our works with analysis on the connection and difference among them. Our methods achieves the second place at the Places2 challenge in ILSVRC 2015, and the first place at the LSUN challenge in CVPR 2016. One question we try to answer with this paper is how scene classification is related to and different from object classification. We hope these investigation and analysis will inspire novel ideas toward the challenging problem of complex scene understanding. |
| 语种 | 英语 |
| 源URL | [http://ir.siat.ac.cn:8080/handle/172644/11772]  |
| 专题 | 深圳先进技术研究院_集成所 |
| 作者单位 | 2017 |
| 推荐引用方式 GB/T 7714 | Yu Qiao,Limin Wang,Sheng Guo,et al. Good Practice on Deep Scene Classification: from Local Supervision to Knowledge Guided Disambiguation[C]. 见:. 美国. |
入库方式： OAI收割
来源：深圳先进技术研究院
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。