Thank you for your wonderful job.
When I read the paper, I was confused about how is the input feature map factorized into high-frequency and low-frequency branches? Or if the input feature map is just randomly split among channel dimensions, and then use the following operations to extract high-frequency and low-frequency information from Xh1, Xh2, and Xl?
Best,
Yulin
Thank you for your wonderful job.
When I read the paper, I was confused about how is the input feature map factorized into high-frequency and low-frequency branches? Or if the input feature map is just randomly split among channel dimensions, and then use the following operations to extract high-frequency and low-frequency information from Xh1, Xh2, and Xl?
Best,
Yulin