search

TencentYoutuResearch / CrowdCounting-P2PNet

The official codes for the ICCV2021 Oral presentation "Rethinking Counting and Localization in Crowds: A Purely Point-Based Framework"
Other
404 stars 109 forks source link

Softmax Normalization Missing in ClassificationModel #66

Closed Younger330 closed 5 months ago

Younger330 commented 7 months ago

It seems that you have overlooked softmax in the ClassificationModel. However, your paper states, "it outputs the confidence scores with a Softmax normalization." I would like to know which statement is accurate.

KRK11 commented 5 months ago

It seems that you have overlooked softmax in the ClassificationModel. However, your paper states, "it outputs the confidence scores with a Softmax normalization." I would like to know which statement is accurate.

The code has softmax for ClassificationModel, which can be found in matcher.py.

Younger330 commented 5 months ago

It seems that you have overlooked softmax in the ClassificationModel. However, your paper states, "it outputs the confidence scores with a Softmax normalization." I would like to know which statement is accurate.

The code has softmax for ClassificationModel, which can be found in matcher.py.

Thanks, it's really helpful.