CrossEntropy
commented
2 months ago Hi, D! @D-VVY, I think the problem with this step is the 1x1 convolution operation. To be precise, it is the squeeze operation that turns the four-dimensional tensor into a two-dimensional one. You can modify the following code to try it out and hope it works!
class conv1x1(nn.Module):
def __init__(self, in_planes, out_planes, stride=1):
super(conv1x1, self).__init__()
self.conv = nn.Conv2d(in_planes, out_planes, kernel_size=1, stride=stride,
padding=0, bias=False)
self.bn = nn.BatchNorm2d(out_planes)
self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
def forward(self, feature):
output = self.conv(feature)
output = self.bn(output)
output = self.avgpool(output)
# output = output.squeeze() # old operation
# new operation
bs, c, h, w = output.shape
output = torch.reshape(output, (bs, c * h * w))
return output
=====================>> (End) Traning params << ======================= ===> Read data <=== -- [Read Data]: Total num: 29538 -- [Read Data]: Source: ['./data/Norm/label_class/train2.txt', './data/Norm/label_class/train3.txt'] ===> Model building <=== Downloading: "https://download.pytorch.org/models/resnet18-5c106cde.pth" to /root/.cache/torch/hub/checkpoints/resnet18-5c106cde.pth 100.0% ===> optimizer building <=== ===> Training <=== Traceback (most recent call last): File "trainer/leave.py", line 184, in
main(configi)
File "trainer/leave.py", line 133, in main
loss, losslist = net.loss(data, anno)
File "/root/autodl-tmp/IVGaze-1/model.py", line 96, in loss
gaze, , loss_gaze_o, loss_gaze_n = self.forward(x_in)
File "/root/autodl-tmp/IVGaze-1/model.py", line 58, in forward
feature_o, feature_list_o= self.borigin(x_in['origin_face'])
File "/root/miniconda3/envs/gaze/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1194, in _call_impl
return forward_call(*input, *kwargs)
File "/root/autodl-tmp/IVGaze-1/IVModule.py", line 400, in forward
feature = self.transformer(feature, self.outFeatureNum)
File "/root/miniconda3/envs/gaze/lib/python3.8/site-packages/torch/nn/modules/module.py", line 1194, in _call_impl
return forward_call(input, **kwargs)
File "/root/autodl-tmp/IVGaze-1/IVModule.py", line 342, in forward
feature_in = torch.cat([cls, feature], 0)
RuntimeError: Tensors must have same number of dimensions: got 3 and 2
When I run it locally, it shows a dimension mismatch. Has anyone encountered the same problem? Can anyone give me some advice? Thanks!