focal-loss icon indicating copy to clipboard operation
focal-loss copied to clipboard

Focal loss for Dense Object Detection

Results 24 focal-loss issues
Sort by recently updated
recently updated
newest added

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?

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...