Problem with Benchmark notebook

STAC Overflow: Map Floodwater from Radar Imagery
1

Hi All,
I was trying to the run the benchmark code in google colab, But I was run into problem . could please anyone help me to solve this.
here my error.

   TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.
2

Tensors on CUDA cannot be converted to numpy directly. You need to manually move it to CPU first.
The solution is exactly what is error description says: x.detach().cpu().numpy().

3

what worked for me was changing true and preds to preds.to(‘cpu’).numpy() and true.to(‘cpu’).numpy()

1 Like
4

Thanks for the solution polymath.

5

@suresh12 were you able to solve it? I am still facing the issue after changing the line
return preds.detach().numpy().squeeze() to return preds.to(‘cpu’).detach().numpy().squeeze()

Do we have to change somewhere else to? can you share the code snippet which you changed
@polymathAB

6

hi
actually you have to use .to(‘cpu’) the first time preds and true are stored as numpy tensors which is in the following line:

intersection = np.logical_and(true, pred)
union = np.logical_or(true, pred)

1 Like
7

Thanks for the help, solved.

8

Just FYI this has been fixed on the blog post in a recent round of edits. Thanks to @polymathAB for flagging.

1 Like