focal-loss
focal-loss copied to clipboard
Focal loss for Dense Object Detection
hello thanks for implementation sharng is it possible to use your implementation in FasterRCNN in Detectrone2 object detection system? I mean to have Focal loss as loss function in FasterRCNN...
二分类时,α的作用是,让数量少的类别权重增大。 那么多分类时,同理,应该为每个类别设置一个α。 一个统一的α值如何能起到这个作用呢?
@unsky 多谢多谢!
What is your initialization of the detector? Is that exactly the same with original paper to set bias = -log((1-pi)/pi) or use normal softmax for several times like your previous...
我按照论文3.3进行了初始化,但是loss非常大,请问你是如何解决的呢?
你好,感谢你的代码,我很好奇如何在faster rcnn中实现focal loss ,是要把softmax换成focal loss吗?你有提供在faster rcnn中实现它的代码吗?非常感谢。
why CE is not this: ![ce](https://user-images.githubusercontent.com/14250619/41197551-37ef09d4-6c95-11e8-8331-f464722200ca.jpg) by my formula,derivatives is different from yours.
Hi, Is there any implementation of Focal Loss in Keras or an ongoing effort? I really appreciate cooperation in such a work, please e-mail me if so.
In the original RCNN, cfg.TRAIN.BATCH_ROIS rois are selected. Does this implementation use the same setting and just modify the cls loss? Or, much more Rois are selected, following the spirit...