search

maudzung / TTNet-Real-time-Analysis-System-for-Table-Tennis-Pytorch

Unofficial implementation of "TTNet: Real-time temporal and spatial video analysis of table tennis" (CVPR 2020)
https://arxiv.org/pdf/2004.09927.pdf
602 stars 159 forks source link

The fourth ConvBlock of the BallDetection forward() function doesn't use DropOut in file 'models/TTNet.py' #19

Open dozingLee opened 3 years ago

dozingLee commented 3 years ago

Will this greatly affect the TTNet?

` class BallDetection(nn.Module): def init(self, num_frames_sequence, dropout_p): .......

def forward(self, x):
  x = self.relu(self.batchnorm(self.conv1(x)))
  out_block2 = self.convblock2(self.convblock1(x))
  x = self.dropout2d(out_block2)
  out_block3 = self.convblock3(x)
  out_block4 = self.convblock4(out_block3)
  x = self.dropout2d(out_block4)

  # out_block5 = self.convblock5(out_block4) # Original Code
  out_block5 = self.convblock5(x)  # The Code should be this.

  features = self.convblock6(out_block5)
  x = self.dropout2d(features)

  x = x.contiguous().view(x.size(0), -1)
  x = self.dropout1d(self.relu(self.fc1(x)))
  x = self.dropout1d(self.relu(self.fc2(x)))
  out = self.sigmoid(self.fc3(x))

  return out, features, out_block2, out_block3, out_block4, out_block5`