Tensors not converted to CUDA?

[Maenansyn]

Hi,

I'm using test.py in Google Colab and CUDA is available in the runtime (Tesla T4). The builder of the dataparallel module (build_dp) gets the device information as parameters. However, when running the script on gpu I get the following error message:

RuntimeError: Input type (torch.FloatTensor) and weight type (torch.cuda.FloatTensor) should be the same or input should be a MKLDNN tensor and weight is a dense tensor

[/content/mmdetection/mmdet/models/backbones/resnet.py](https://localhost:8080/#) in forward(self, x)
    634             x = self.stem(x)
    635         else:
--> 636             x = self.conv1(x)
    637             x = self.norm1(x)
    638             x = self.relu(x)
/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py in _call_impl(self, *input, **kwargs)

1049         if not (self._backward_hooks or self._forward_hooks or self._forward_pre_hooks or _global_backward_hooks

1050                 or _global_forward_hooks or _global_forward_pre_hooks):

-> 1051             return forward_call(*input, **kwargs)

1052         # Do not call functions when jit is used

1053         full_backward_hooks, non_full_backward_hooks = [], []
/usr/local/lib/python3.7/dist-packages/torch/nn/modules/conv.py in forward(self, input)

441

442     def forward(self, input: Tensor) -> Tensor:

--> 443         return self._conv_forward(input, self.weight, self.bias)

444

445 class Conv3d(_ConvNd):
/usr/local/lib/python3.7/dist-packages/torch/nn/modules/conv.py in _conv_forward(self, input, weight, bias)

438                             _pair(0), self.dilation, self.groups)

439         return F.conv2d(input, weight, bias, self.stride,

--> 440                         self.padding, self.dilation, self.groups)

441

442     def forward(self, input: Tensor) -> Tensor:
RuntimeError: Input type (torch.FloatTensor) and weight type (torch.cuda.FloatTensor) should be the same or input should be a MKLDNN tensor and weight is a dense tensor

torch.cuda.is_available() returns True, isinstance(img, torch.cuda.FloatTensor) during the end of the test-transforms pipeline (adapted DefaultFormatBundle.transform) returns False nonetheless.

Any tip how to fix this or any clue about the reason? When trying to convert the tensor at that point with Tensor.to(device) I got a RuntimeError regarding multiprocessing despite single gpu (calling single_gpu_test):

RuntimeError: Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you must use the 'spawn' start method