search

cvlab-stonybrook / DM-Count

Code for NeurIPS 2020 paper: Distribution Matching for Crowd Counting.
MIT License
218 stars 52 forks source link

Why not use more advanced backbone #15

Closed hhk123 closed 3 years ago

hhk123 commented 3 years ago

Excellent work! But why not use more advanced backbone, such as resnet or CSPNet et.? Is that because of in order to prove the progresiveness of the loss designed?

Boyu-Wang commented 3 years ago

Hi, Thank you for your attention to our work! For fair comparisons, we use the same backbone network as other losses. You are welcome to replace the backbone with other state-of-the-art networks.

hhk123 commented 3 years ago

Hi, Thank you for your attention to our work! For fair comparisons, we use the same backbone network as other losses. You are welcome to replace the backbone with other state-of-the-art networks.

I see.Thank you and happy Chiness new year!

Hasan-Torabi commented 3 years ago

@hhk123 Hi, Good idea. Have you tried using other backbones?