search

Booooooooooo / AECRNet-MindSpore

MindSpore implementation of Contrastive Learning for Compact Single Image Dehazing, CVPR2021
64 stars 8 forks source link

关于model.py中Dehaze模型的实现细节问题？ #4

Open foreverfruit opened 3 years ago

foreverfruit commented 3 years ago 
def __init__(self, input_nc, output_nc, ngf=64):
    ...
    self.block = DehazeBlock(default_conv, ngf*4, 3)
    ...

def construct(self, x):
    ...
    x1 = self.block(x_down3)
    x2 = self.block(x1)

    x3 = self.block(x2)
    x4 = self.block(x3)

    x5 = self.block(x4)
    x6 = self.block(x5)
    ....

这样写是不是会造成网络中只有一个DehazeBlock，而不是论文中所说的6个？？

TaoWangzj commented 3 years ago

Same question, whether the code has some mistake? Looking forward to your reply.

wpf535236337 commented 2 years ago

好问题，按照原文“Note that we significantly reduce the number of FA blocks by only using 6 FA blocks (vs. 57 FA blocks in FFA-Net)” FFA-Net image 确实不应该是参数共享？