Transcription of Focal Loss for Dense Object Detection
{{id}} {{{paragraph}}}
Focal Loss for Dense Object DetectionTsung-Yi Lin Priya Goyal Ross Girshick Kaiming He Piotr Doll arFacebook AI Research (FAIR) of ground truth class012345loss = 0 = = 1 = 2 = 5well-classi edexampleswell-classi edexamplesCE(pt) = log(pt)FL(pt) = (1 pt) log(pt)Figure 1. We propose a novel loss we term theFocal Lossthatadds a factor(1 pt) to the standard cross entropy >0reduces the relative loss for well-classified examples(pt> .5), putting more focus on hard, misclassified examples. Asour experiments will demonstrate, the proposed Focal loss enablestraining highly accurate Dense Object detectors in the presence ofvast numbers of easy background highest accuracy Object detectors to date are basedon a two-stage approach popularized by R-CNN, where aclassifier is applied to asparseset of candidate Object lo-cations.
This paper pushes the envelop further: we present a one-stage object detector that, for the first time, matches the 12980. state-of-the-art COCO AP of more complex two-stage de-tectors, such as the Feature Pyramid Network (FPN) [19] or Mask R …
Domain:
Source:
Link to this page:
Please notify us if you found a problem with this document:
{{id}} {{{paragraph}}}