Some questions about classification in FashionNet

[zuoxiang95]

hello @liuziwei7, I have been reading your paper Deepfashion. It is a bit confused about the loss function of attribute classification in the paper. The paper says w_pos and w_neg are two coefficients, determined by the ratio of the numbers of positive and negative samples in the dataset. My understanding is like this:

                       w_pos = num_positve / num_dataset
                       w_neg = 1 - w_pos

Is my understanding wrong? Looking forward to your reply, thank you very much!