Can someone help me to understand "self.fc = nn.Linear(176, 196)" how 176 input feature has been choose

[sainisanjay]

Can someone help me to understand self.fc = nn.Linear(176, 196) how 176 input feature has been choose in PFLDInference network.

[Sridhar98]

I have the same question. Do you have any idea about this @sainisanjay ?

[CrossEntropy]

That's because when you stack three multi-scale features together, you get a new tensor which shape is (bs, 176). Please see the code in models/pfld.py @sainisanjay @Sridhar98