There are some discrepancies between your paper and code.

[liunangithub]

``class DenseNetBlock(nn.Module):
def init(self, inplanes, growth_rate, drop_prob=0.0):
super().init()
self.dense_block = ### ### nn.Sequential(
Conv2d(inplanes, 4 * growth_rate,
kernel_size=1, stride=1, padding=0, drop_prob=drop_prob),
Conv2d(4 * growth_rate, growth_rate,
kernel_size=3, stride=1, padding=1, drop_prob=drop_prob, act=None)
)

def forward(self, input):
    y = self.dense_block(input)
    y = torch.cat([input, y], dim=1)
    return y

in the part , I think it may miss a ELU function between two COnv2d models.Am I right?