Open joihn opened 7 months ago
hardnet v1 does not have conv2d bias. ptsemseg/models/hardnet.py:11
ptsemseg/models/hardnet.py:11
self.add_module('conv', nn.Conv2d(in_channels, out_channels, kernel_size=kernel, stride=stride, padding=kernel//2, bias = False))
However, hardnetv2, which use catconv2d does have conv2d bias CatConv2d/catconv2d.py:39
CatConv2d/catconv2d.py:39
self.bias = nn.Parameter(torch.Tensor(out_channels))
Why does catconv2d has bias ? isn't it useless ?
hardnet v1 does not have conv2d bias.
ptsemseg/models/hardnet.py:11
However, hardnetv2, which use catconv2d does have conv2d bias
CatConv2d/catconv2d.py:39
Why does catconv2d has bias ? isn't it useless ?