search

IIGROUP / AHIQ

[CVPRW 2022] Attentions Help CNNs See Better: Attention-based Hybrid Image Quality Assessment Network
MIT License
72 stars 5 forks source link

question about CNN Features #7

Open TimZhang001 opened 1 year ago

TimZhang001 commented 1 year ago

The method uses the features of stage2 of Resnet50 as the result of feature extraction. It's very strange here, why choose stage2, or why not choose one output per stage for concat?