Abstract
Recently developed object detectors employ a convolutional neural network (CNN) by gradually increasing the number of feature layers with a pyramidal shape instead of using a featurized image pyramid. However, the different abstraction levels of CNN feature layers often limit the detection performance, especially on small objects. To overcome this limitation, we propose a CNN-based object detection architecture, referred to as a parallel feature pyramid (FP) network (PFPNet), where the FP is constructed by widening the network width instead of increasing the network depth. First, we adopt spatial pyramid pooling and some additional feature transformations to generate a pool of feature maps with different sizes. In PFPNet, the additional feature transformation is performed in parallel, which yields the feature maps with similar levels of semantic abstraction across the scales. We then resize the elements of the feature pool to a uniform size and aggregate their contextual information to generate each level of the final FP. The experimental results confirmed that PFPNet increases the performance of the latest version of the single-shot multi-box detector (SSD) by mAP of 6.4% AP and especially, 7.8% AP small on the MS-COCO dataset.
| Original language | English |
|---|---|
| Title of host publication | Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings |
| Editors | Vittorio Ferrari, Cristian Sminchisescu, Martial Hebert, Yair Weiss |
| Publisher | Springer Verlag |
| Pages | 239-256 |
| Number of pages | 18 |
| ISBN (Print) | 9783030012274 |
| DOIs | |
| Publication status | Published - 2018 Jan 1 |
| Event | 15th European Conference on Computer Vision, ECCV 2018 - Munich, Germany Duration: 2018 Sep 8 → 2018 Sep 14 |
Publication series
| Name | Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) |
|---|---|
| Volume | 11209 LNCS |
| ISSN (Print) | 0302-9743 |
| ISSN (Electronic) | 1611-3349 |
Other
| Other | 15th European Conference on Computer Vision, ECCV 2018 |
|---|---|
| Country | Germany |
| City | Munich |
| Period | 18/9/8 → 18/9/14 |
Fingerprint
Keywords
- Feature pyramid
- Real-time object detection
ASJC Scopus subject areas
- Theoretical Computer Science
- Computer Science(all)
Cite this
Parallel Feature Pyramid Network for Object Detection. / Kim, Seung Wook; Kook, Hyong Keun; Sun, Jee Young; Kang, Mun Cheon; Ko, Sung-Jea.
Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings. ed. / Vittorio Ferrari; Cristian Sminchisescu; Martial Hebert; Yair Weiss. Springer Verlag, 2018. p. 239-256 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 11209 LNCS).Research output: Chapter in Book/Report/Conference proceeding › Conference contribution
}
TY - GEN
T1 - Parallel Feature Pyramid Network for Object Detection
AU - Kim, Seung Wook
AU - Kook, Hyong Keun
AU - Sun, Jee Young
AU - Kang, Mun Cheon
AU - Ko, Sung-Jea
PY - 2018/1/1
Y1 - 2018/1/1
N2 - Recently developed object detectors employ a convolutional neural network (CNN) by gradually increasing the number of feature layers with a pyramidal shape instead of using a featurized image pyramid. However, the different abstraction levels of CNN feature layers often limit the detection performance, especially on small objects. To overcome this limitation, we propose a CNN-based object detection architecture, referred to as a parallel feature pyramid (FP) network (PFPNet), where the FP is constructed by widening the network width instead of increasing the network depth. First, we adopt spatial pyramid pooling and some additional feature transformations to generate a pool of feature maps with different sizes. In PFPNet, the additional feature transformation is performed in parallel, which yields the feature maps with similar levels of semantic abstraction across the scales. We then resize the elements of the feature pool to a uniform size and aggregate their contextual information to generate each level of the final FP. The experimental results confirmed that PFPNet increases the performance of the latest version of the single-shot multi-box detector (SSD) by mAP of 6.4% AP and especially, 7.8% AP small on the MS-COCO dataset.
AB - Recently developed object detectors employ a convolutional neural network (CNN) by gradually increasing the number of feature layers with a pyramidal shape instead of using a featurized image pyramid. However, the different abstraction levels of CNN feature layers often limit the detection performance, especially on small objects. To overcome this limitation, we propose a CNN-based object detection architecture, referred to as a parallel feature pyramid (FP) network (PFPNet), where the FP is constructed by widening the network width instead of increasing the network depth. First, we adopt spatial pyramid pooling and some additional feature transformations to generate a pool of feature maps with different sizes. In PFPNet, the additional feature transformation is performed in parallel, which yields the feature maps with similar levels of semantic abstraction across the scales. We then resize the elements of the feature pool to a uniform size and aggregate their contextual information to generate each level of the final FP. The experimental results confirmed that PFPNet increases the performance of the latest version of the single-shot multi-box detector (SSD) by mAP of 6.4% AP and especially, 7.8% AP small on the MS-COCO dataset.
KW - Feature pyramid
KW - Real-time object detection
UR - http://www.scopus.com/inward/record.url?scp=85055106425&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85055106425&partnerID=8YFLogxK
U2 - 10.1007/978-3-030-01228-1_15
DO - 10.1007/978-3-030-01228-1_15
M3 - Conference contribution
AN - SCOPUS:85055106425
SN - 9783030012274
T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
SP - 239
EP - 256
BT - Computer Vision – ECCV 2018 - 15th European Conference, 2018, Proceedings
A2 - Ferrari, Vittorio
A2 - Sminchisescu, Cristian
A2 - Hebert, Martial
A2 - Weiss, Yair
PB - Springer Verlag
ER -