Conv2d value cannot be converted to type uint8_t without overflow

1

I’m passing a torch.Tensor with a dtype of torch.uint8 to an nn.Conv2d module and it is giving the error

RuntimeError: value cannot be converted to type uint8_t without overflow: -0.0344873

My conv2d is defined as self.conv1 = nn.Conv2d(3, 6, 5) . The error comes in my forward method when I pass the tensor to the module like self.conv1(x) . The tensor has shape (4, 3, 480, 640). I’m not sure how to fix this. Here is the stack trace

Traceback (most recent call last):

File “cnn.py”, line 54, in 

outputs = net(inputs)

File “/Users/my_repos/venv_projc/lib/python3.7/site-packages/torch/nn/modules/module.py”, line 532, in call

result = self.forward(*input, **kwargs)

File “cnn.py”, line 24, in forward

test = self.conv1(x)

File “/Users/my_repos/venv_projc/lib/python3.7/site-packages/torch/nn/modules/module.py”, line 532, in call

result = self.forward(*input, **kwargs)

File “/Users/my_repos/venv_projc/lib/python3.7/site-packages/torch/nn/modules/conv.py”, line 345, in forward

return self.conv2d_forward(input, self.weight)

File “/Users/my_repos/venv_projc/lib/python3.7/site-packages/torch/nn/modules/conv.py”, line 342, in conv2d_forward

self.padding, self.dilation, self.groups)

RuntimeError: value cannot be converted to type uint8_t without overflow: -0.0344873

2

Converting the tensor to a float seemed to fix it self.conv1(x.float())

3

Hi,

Yes the error message is not very clear but the regular conv2d only works for float and double numbers.

4

so how can I do inference with uint8 ? I have saved integer weights and I want to evaluate my model with that, is it possible?