Skip to content

Latest commit

 

History

History
108 lines (28 loc) · 1.23 KB

File metadata and controls

108 lines (28 loc) · 1.23 KB

图像和视觉相关的loss函数

Low-level 任务常用loss

  • L1 / L2 loss

最常用loss函数,属于 pixel loss,即直接拟合目标图像。只能针对对齐的pair数据。

  • Charbonnier loss

ref: Fast and Accurate Image Super-Resolution with Deep Laplacian Pyramid Networks section 3.3

  • TV loss(total variation)

  • content loss(vgg loss)

约束内容一致性,可以通过pretrained VGG等网络提取特征。

  • texture loss

约束纹理风格一致性

  • GAN loss

生成结果符合目标图像分布

分割任务常用loss(从mmseg梳理)

  • Cross-Entropy loss

  • IOU loss

  • Dice loss

  • Focal loss

  • Lovasz loss

  • Tversky loss

目标检测/实例分割常用loss(从mmdet梳理)

  • AE loss(Associative Embedding Loss)

  • Balanced L1 loss

ref:Libra R-CNN: Towards Balanced Learning for Object Detection

  • PISA loss(PrIme Sample Attention (PISA) )

ref: Prime Sample Attention in Object Detection

  • Seesaw loss

ref: [Seesaw Loss for Long-Tailed Instance Segmentation](arXiv: https://arxiv.org/abs/2008.10032)