Focal Loss for Dense Object Detection

Authors: Lin, T. Y., Goyal, P., Girshick, R., He, K., & Dollár, P. Year: 2017 Venue: ICCV ArXiv: https://arxiv.org/abs/1708.02002

Summary

RetinaNet introduces focal loss to address class imbalance in one-stage detectors. It achieves better accuracy than Faster R-CNN while maintaining real-time performance.

Key Concepts

• Focal loss for hard example mining
• Dense object detection
• Feature pyramids (FPN)
• Addressing class imbalance
• One-stage detector improvements

Impact

RetinaNet proved one-stage detectors could match two-stage accuracy by properly handling class imbalance. Focal loss became widely adopted beyond detection.

Related Papers

• YOLO: Real-Time Detection
• SSD: Single Shot Detection
• Faster R-CNN: Two-Stage Detection

---

Citation:

@inproceedings{lin2017focal,
  title={Focal Loss for Dense Object Detection},
  author={Lin, Tsung-Yi and Goyal, Priya and Girshick, Ross and He, Kaiming and Dollár, Piotr},
  booktitle={IEEE International Conference on Computer Vision},
  year={2017}
}