NVlabs / AL-MDN

Official pytorch implementation of Active Learning for deep object detection via probabilistic modeling (ICCV 2021)
https://openaccess.thecvf.com/content/ICCV2021/html/Choi_Active_Learning_for_Deep_Object_Detection_via_Probabilistic_Modeling_ICCV_2021_paper.html
Other
167 stars 23 forks source link

How can I apply this algorithm for detector with focal loss #1

Closed lichengwei-code closed 3 years ago

lichengwei-code commented 3 years ago

Hi, I am very interested in your work. So I wanna apply this algorithm for my work. As a commonly used loss function focal loss, the output of clssification is different from the cross entropy loss. The classification output layer num is equal to the class num, not class num add 1, So I wanna know how to change the loss function in this paper for applying focal loss output. Thanks a lot.

jwchoi384 commented 3 years ago

Hello, I have known that focal loss in their paper (https://arxiv.org/pdf/1708.02002.pdf) is based on a cross-entropy loss with a modulating factor. I think you need to add a modulating factor that takes into account Gaussian mixing parameters to our proposed loss function.

Shanyaodedanshen commented 3 years ago

Hello, I have known that focal loss in their paper (https://arxiv.org/pdf/1708.02002.pdf) is based on a cross-entropy loss with a modulating factor. I think you need to add a modulating factor that takes into account Gaussian mixing parameters to our proposed loss function.

"add a modulating factor that takes into account Gaussian mixing parameters to our proposed loss function." Yes, I agree with it. It is also recommended that the project "pod-compare", "probdet" in github.