SSD: Single Shot MultiBox Detector

Authors: Liu, W., Anguelov, D., Erhan, D., et al. Year: 2016 Venue: ECCV ArXiv: https://arxiv.org/abs/1512.02325

Summary

SSD improves YOLO by using multi-scale feature maps for detection. It achieves both speed and accuracy, becoming one of the most popular real-time detectors.

Key Concepts

• Multi-scale feature maps
• Default boxes at multiple scales
• Fully convolutional architecture
• Fast inference (59 FPS)
• Strong accuracy-speed balance

Impact

SSD became the de facto standard for efficient object detection, balancing speed and accuracy better than YOLO or Faster R-CNN.

Related Papers

• YOLO: Real-Time Object Detection
• Faster R-CNN: Towards Real-Time Detection
• RetinaNet: Focal Loss for Dense Detection

---

Citation:

@inproceedings{liu2016ssd,
  title={SSD: Single shot multibox detector},
  author={Liu, Wei and Anguelov, Dragomir and Erhan, Dumitru and Szegedy, Christian and others},
  booktitle={European Conference on Computer Vision},
  year={2016}
}