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

fastai / course-v3

The 3rd edition of course.fast.ai

Apache License 2.0

4.91k stars 3.57k forks source link

Copied the tensors on cpu as in the below function in fastai dl2 applications cyclegan.ipynb and cyclegan_ws.ipynb :

def show_best(fnames, losses, gen, n=8):
    sort_idx = losses.argsort().cpu()
    _,axs = plt.subplots(n//2, 4, figsize=(12,2*n))
    xb = get_batch(fnames[sort_idx][:n], tfms, size=128)
    with torch.no_grad():
        fakes = gen(xb)
    xb,fakes = (1+xb.cpu())/2,(1+fakes.cpu())/2
    for i in range(n):
        axs.flatten()[2*i].imshow(xb[i].permute(1,2,0))
        axs.flatten()[2*i].axis('off')
        axs.flatten()[2*i+1].imshow(fakes[i].permute(1,2,0))
        axs.flatten()[2*i+1].set_title(losses[sort_idx][i].item())
        axs.flatten()[2*i+1].axis('off')

fastai / course-v3

TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first. #519