Tong WU

it's the formula of focal loss, and "labels" should be reserved.  notice that y' = 1 / (1 + torch.exp(-1.0 * logits)), and both logits and labels are matrixs....

Hi, sorry for the late reply! To apply it in your own datasets, you may: 1. Merge your datasets to this project: define the `dataset Class` following the coco/voc/lvis style...

Hi @Jacobew, Notice the `if-else` condition in that part of the function: the parameter `reduction` is used [here](https://github.com/wutong16/DistributionBalancedLoss/blob/a3ecaa9021a920fcce9fdafbd7d83b51bf526af8/mllt/models/losses/resample_loss.py#L127) in the `else` branch. While in the `if self.focal` branch, the `reduction`...

Hi @DinSangrasi! Sorry for the delayed reply. We implement our loss [here](https://github.com/wutong16/Adversarial_Long-Tail/blob/main/methods/losses_inclass.py#L280) in the code following Eqn. (9), where the margin is defined [here](https://github.com/wutong16/Adversarial_Long-Tail/blob/main/methods/losses_inclass.py#L266) following Eqn. (8) and the bias...

Hi @stasinak, I'm not aware of the normalization method either...