| | |
|---|
| | | |
|---|
| | | return pre_yaw, pre_pitch, pre_roll, angles, sr_output |
|---|
| | | |
|---|
| | | class Hopenet_LSTM(nn.Module): |
|---|
| | | class Hopenet_new(nn.Module): |
|---|
| | | # This is just Hopenet with 3 output layers for yaw, pitch and roll. |
|---|
| | | def __init__(self, block, layers, num_bins): |
|---|
| | | self.inplanes = 64 |
|---|
| | | super(Hopenet_LSTM, self).__init__() |
|---|
| | | super(Hopenet_new, self).__init__() |
|---|
| | | self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3, |
|---|
| | | bias=False) |
|---|
| | | self.bn1 = nn.BatchNorm2d(64) |
|---|
| | |
|---|
| | | self.fc_roll = nn.Linear(512 * block.expansion, num_bins) |
|---|
| | | |
|---|
| | | self.softmax = nn.Softmax() |
|---|
| | | self.fc_finetune = nn.Linear(512 * block.expansion + 3, 3) |
|---|
| | | self.fc_finetune_new = nn.Linear(512 * block.expansion + 256 * block.expansion + 3, 3) |
|---|
| | | self.conv1x1 = nn.Conv2d(1024, 64, kernel_size = 1, stride = 1, bias=False) |
|---|
| | | self.maxpool_interm = nn.MaxPool2d(kernel_size=5, stride=3, padding=1) |
|---|
| | | |
|---|
| | | self.idx_tensor = Variable(torch.FloatTensor(range(66))).cuda() |
|---|
| | | |
|---|
| | | self.lstm = nn.LSTM(512 * block.expansion + 3, 256 * block.expansion, 2, batch_first=True) |
|---|
| | | self.fc_lstm = nn.Linear(256 * block.expansion, 3) |
|---|
| | | |
|---|
| | | self.block_expansion = block.expansion |
|---|
| | | |
|---|
| | | for m in self.modules(): |
|---|
| | | if isinstance(m, nn.Conv2d): |
|---|
| | |
|---|
| | | return nn.Sequential(*layers) |
|---|
| | | |
|---|
| | | def forward(self, x): |
|---|
| | | |
|---|
| | | x = self.conv1(x) |
|---|
| | | x = self.bn1(x) |
|---|
| | | x = self.relu(x) |
|---|
| | |
|---|
| | | x = self.layer1(x) |
|---|
| | | x = self.layer2(x) |
|---|
| | | x = self.layer3(x) |
|---|
| | | x_interm = self.conv1x1(x) |
|---|
| | | x_interm = self.relu(x_interm) |
|---|
| | | x_interm = self.maxpool_interm(x_interm) |
|---|
| | | x_interm = x_interm.view(x_interm.size(0), -1) |
|---|
| | | |
|---|
| | | x = self.layer4(x) |
|---|
| | | |
|---|
| | | x = self.avgpool(x) |
|---|
| | |
|---|
| | | pre_pitch = self.fc_pitch(x) |
|---|
| | | pre_roll = self.fc_roll(x) |
|---|
| | | |
|---|
| | | # Yaw, pitch, roll |
|---|
| | | yaw = self.softmax(pre_yaw) |
|---|
| | | yaw = Variable(torch.sum(yaw.data * self.idx_tensor.data, 1), requires_grad=True) * 3 - 99 |
|---|
| | | pitch = self.softmax(pre_pitch) |
|---|
| | |
|---|
| | | roll = roll.view(roll.size(0), 1) |
|---|
| | | preangles = torch.cat([yaw, pitch, roll], 1) |
|---|
| | | |
|---|
| | | residuals, _ = self.lstm(torch.cat((preangles, x), 1), (h0, c0)) |
|---|
| | | residuals = self.fc_lstm(residuals[:, -1, :]) |
|---|
| | | # angles predicts the residual |
|---|
| | | residuals = self.fc_finetune_new(torch.cat((preangles, x_interm, x), 1)) |
|---|
| | | final_angles = preangles + residuals |
|---|
| | | |
|---|
| | | return pre_yaw, pre_pitch, pre_roll, preangles, final_angles |
|---|
