yjxiong / tsn-pytorch

Temporal Segment Networks (TSN) in PyTorch
BSD 2-Clause "Simplified" License
1.07k stars 308 forks source link

LOSS function is different from the paper #80

Open fmthoker opened 5 years ago

fmthoker commented 5 years ago

@yjxiong Can you explain why there is a different loss function mentioned in the paper as opposed to the normal cross entropy? From the implementation, I can see you have used normal cross entropy over average predictions from the K shared networks. Am I missing something here?