Enabling Deep Residual Networks for Weakly Supervised Object Detection

Shen，Yunhang1; Ji，Rongrong1; Wang，Yan2; Chen，Zhiwei1; Zheng，Feng3; Huang，Feiyue4; Wu，Yunsheng4

10.1007/978-3-030-58598-3_8

2020

0302-9743

1611-3349

LECT NOTES ARTIF INT

12353 LNCS

118-136

Weakly supervised object detection (WSOD) has attracted extensive research attention due to its great flexibility of exploiting large-scale image-level annotation for detector training. Whilst deep residual networks such as ResNet and DenseNet have become the standard backbones for many computer vision tasks, the cutting-edge WSOD methods still rely on plain networks, e.g., VGG, as backbones. It is indeed not trivial to employ deep residual networks for WSOD, which even shows significant deterioration of detection accuracy and non-convergence. In this paper, we discover the intrinsic root with sophisticated analysis and propose a sequence of design principles to take full advantages of deep residual learning for WSOD from the perspectives of adding redundancy, improving robustness and aligning features. First, a redundant adaptation neck is key for effective object instance localization and discriminative feature learning. Second, small-kernel convolutions and MaxPool down-samplings help improve the robustness of information flow, which gives finer object boundaries and make the detector more sensitivity to small objects. Third, dilated convolution is essential to align the proposal features and exploit diverse local information by extracting high-resolution feature maps. Extensive experiments show that the proposed principles enable deep residual networks to establishes new state-of-the-arts on PASCAL VOC and MS COCO.

其他

英语

EI

20205009617791

Arts computing ; Object recognition ; Deep learning ; Deterioration ; Computer vision ; Convolution

Ergonomics and Human Factors Engineering:461.4 ; Information Theory and Signal Processing:716.1 ; Data Processing and Image Processing:723.2 ; Computer Applications:723.5 ; Vision:741.2 ; Materials Science:951

2-s2.0-85097395636

Scopus

1.Media Analytics and Computing Lab,Department of Artificial Intelligence,School of Informatics,Xiamen University,Xiamen,361005,China
2.Pinterest,San Francisco,United States
3.CSE,Southern University of Science and Technology,Shenzhen,China
4.Tencent Youtu Lab,Tencent Technology (Shanghai) Co.,Ltd.,Shanghai,China

Shen，Yunhang,Ji，Rongrong,Wang，Yan,et al. Enabling Deep Residual Networks for Weakly Supervised Object Detection[C],2020:118-136.