【已解决】TypeError: can‘t convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the host

问题

在训练Pytorch模型的时候，报错

Traceback (most recent call last):
  File "F:\project\modelscope-master\modelscope-master\modelscope.py", line 17, in 
    result = inpainting(input)
  File "F:\project\modelscope-master\modelscope-master\modelscope\pipelines\base.py", line 219, in __call__
    output = self._process_single(input, *args, **kwargs)
  File "F:\project\modelscope-master\modelscope-master\modelscope\pipelines\base.py", line 247, in _process_single
    out = self.preprocess(input, **preprocess_params)
  File "F:\project\modelscope-master\modelscope-master\modelscope\pipelines\cv\image_inpainting_pipeline.py", line 107, in preprocess
    result = self.perform_inference(result)
  File "F:\project\modelscope-master\modelscope-master\modelscope\pipelines\cv\image_inpainting_pipeline.py", line 119, in perform_inference
    cur_res = refine_predict(
  File "F:\project\modelscope-master\modelscope-master\modelscope\models\cv\image_inpainting\refinement.py", line 385, in refine_predict
    image_inpainted = _infer(image, mask, forward_front, forward_rears,
  File "F:\project\modelscope-master\modelscope-master\modelscope\models\cv\image_inpainting\refinement.py", line 214, in _infer
    mask_downscaled = _erode_mask(mask_downscaled, ekernel=ekernel)
  File "F:\project\modelscope-master\modelscope-master\modelscope\models\cv\image_inpainting\refinement.py", line 109, in _erode_mask
    mask = erosion(mask, ekernel)
  File "I:\conda\envs\lama\lib\site-packages\skimage\morphology\misc.py", line 39, in func_out
    return func(image, selem=selem, *args, **kwargs)
  File "I:\conda\envs\lama\lib\site-packages\skimage\morphology\grey.py", line 181, in erosion
    selem = np.array(selem)
  File "I:\conda\envs\lama\lib\site-packages\torch\_tensor.py", line 757, in __array__
    return self.numpy()
TypeError: can't convert cuda:0 device type tensor to numpy. Use Tensor.cpu() to copy the tensor to host memory first.

问题分析

如果想把CUDA tensor格式的数据改成numpy时，需要先将其转换成cpu float-tensor随后再转到numpy格式。 numpy不能读取CUDA tensor 需要将它转化为 CPU tensor

解决方法

将报错代码self.numpy()改为self.cpu().numpy()即可

补充说明

之前报错的代码是在101服务器上，创建的虚拟环境中Python=3.7，在跑实验的时候出现报错；今天在100服务器上跑同样的实验没有报错，经查在100服务器上创建的虚拟环境中Python=3.8；目测是Python版本的缘故，看来还是python3.8好用一些