</> Code with Papers by Syntology

Methods › General › Loss Functions › Focal Loss

Focal Loss

461 papers tagged archive 2025-07-28

Introduced by Tsung-Yi Lin et al. in Focal Loss for Dense Object Detection

archive 2025-07-28 Description, source and code snippet are the archive's method entry.

A Focal Loss function addresses class imbalance during training in tasks like object detection. Focal loss applies a modulating term to the cross entropy loss in order to focus learning on hard misclassified examples. It is a dynamically scaled cross entropy loss, where the scaling factor decays to zero as confidence in the correct class increases. Intuitively, this scaling factor can automatically down-weight the contribution of easy examples during training and rapidly focus the model on hard examples.

Formally, the Focal Loss adds a factor (1 - pₜ)^γ to the standard cross entropy criterion. Setting γ>0 reduces the relative loss for well-classified examples (pₜ>.5), putting more focus on hard, misclassified examples. Here there is tunable focusing parameter γ≥0.

FL(pₜ) = - (1 - pₜ)^γ log(pₜ)

PaperSourceSee Code · clcarwin/focal_loss_pytorch

Papers archive 2025-07-28

30 shown of 461, newest first. Repository counts are the archive's code-links table. A Syntology line states what Syntology ran from that paper's harvested code; it is per sample and not a correctness claim.

Tasks archive 2025-07-28

20 shown of 341 tasks the archive attaches to papers tagged with this method, by distinct papers. A task without a page in the catalog is plain text.

TaskPapers
Object Detection202
object-detection186
Object121
Semantic Segmentation46
Image Classification33
Segmentation33
General Classification25
image-classification25
Knowledge Distillation23
Classification21
Instance Segmentation21
Data Augmentation20
Deep Learning19
Decoder18
Autonomous Driving16
Transfer Learning16
Image Segmentation13
Pedestrian Detection13
regression12
GPU11

Usage over time archive 2025-07-28

Papers per year tagged with Focal Loss: 2017 to 2025, peak 84 84 0 2017: 5 papers 2017 2018: 26 papers 2018 2019: 68 papers 2019 2020: 66 papers 2020 2021: 80 papers 2021 2022: 67 papers 2022 2023: 84 papers 2023 2024: 48 papers 2024 2025: 17 papers 2025
Papers per year the archive tags with this method, by the paper's archive date (461 dated). Bars are counts, not a trend claim.

Components: the archive holds no method-to-method composition, so PwC's Components table cannot be rebuilt; the Papers list carries no Results column for the same reason (the archive does not join its leaderboard rows to method tags).

Categories archive 2025-07-28

Loss Functions

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections