Skip to main content

> ML_LITERATURE // LIN-2017-FOCAL-LOSS-FOR-DENSE-OBJECT-DETECTION_v1.0

Focal Loss for Dense Object Detection (RetinaNet)

Tsung-Yi Lin, Priya Goyal, Ross Girshick, Kaiming He, Piotr Dollár · IEEE International Conference on Computer Vision (ICCV) (2017)

algorithm2017industry-standardthirdPartyReproduced
DOI: 10.1109/ICCV.2017.324Back to Literature

Principal Contribution

Designed Focal Loss FL(p_t) = -alpha_t * (1 - p_t)^gamma * log(p_t), dynamically down-weighting easy background examples to solve severe class imbalance.

Operational Relevance

Serves as canonical technical reference for implementing task-object-detection, task-imbalanced-classification in production systems.

Assumptions

  • Underlying data distribution satisfies empirical consistency and regularity assumptions across training domains

Limitations

  • Scaling characteristics and accuracy depend on hardware architecture, parameter scale, and dataset quality

Connected Algorithms, Architectures & Tools

Related Algorithms:
Simple Framework for Contrastive Learning (SimCLR)
Related Architectures:
Implementing Libraries: