search

zhiyuanyou / SAFECount

[WACV 2023] Few-shot Object Counting with Similarity-Aware Feature Enhancement
Other
124 stars 22 forks source link

About support image #33

Closed junjun1023 closed 6 months ago

junjun1023 commented 7 months ago

I appreciate your excellent work.

What confuses me is where the support images are read and fed to network and extract feature?

Thank you for the response.

zhiyuanyou commented 7 months ago

In FSC-147 dataset, the support images are crop from the original image. Thus we only input the bounding boxes to the networks, as shown in https://github.com/zhiyuanyou/SAFECount/blob/main/models/safecount.py#L56.

zhiyuanyou commented 7 months ago

In other datasets, the support images are used in https://github.com/zhiyuanyou/SAFECount/blob/main/models/safecount_exemplar.py#L58 and https://github.com/zhiyuanyou/SAFECount/blob/main/models/safecount_exemplar.py#L59.