一、源码

def np2Tensor(l, rgb_range):    def _np2Tensor(img):        # if img.shape[2] == 3: # for opencv imread        #     img = img[:, :, [2, 1, 0]]        np_transpose = np.ascontiguousarray(img.transpose((2, 0, 1)), dtype=np.float32)        tensor = torch.from_numpy(np_transpose).float()        tensor.mul_(rgb_range / 255.)        return tensor    return [_np2Tensor(_l) for _l in l]

二、报错

在复现SRFBN的时候遇到了一个报错，这里记录一下解决的过程，以为备忘！！！

File "D:\source code\SRFBN_CVPR19\data\LRHR_dataset.py", line 41, in __getitem__    lr_tensor, hr_tensor = common.np2Tensor([lr, hr], self.opt['rgb_range'])  File "D:\source code\SRFBN_CVPR19\data\common.py", line 119, in np2Tensor    return [_np2Tensor(_l) for _l in l]  File "D:\source code\SRFBN_CVPR19\data\common.py", line 119, in 
   
        return [_np2Tensor(_l) for _l in l]  File "D:\source code\SRFBN_CVPR19\data\common.py", line 114, in _np2Tensor    tensor = torch.from_numpy(np_transpose).float()ValueError: some of the strides of a given numpy array are negative. This is currently not supported, but will be added in future releases.

三、原因

这个原因是因为程序中操作的numpy中有使用负索引的情况

四、解决方案

**修改之处：**在np_transpose = np.ascontiguousarray(img.transpose((2, 0, 1)), dtype=np.float32)中指定了数据类型。

def np2Tensor(l, rgb_range):    def _np2Tensor(img):        # if img.shape[2] == 3: # for opencv imread        #     img = img[:, :, [2, 1, 0]]        np_transpose = np.ascontiguousarray(img.transpose((2, 0, 1)), dtype=np.float32)        tensor = torch.from_numpy(np_transpose).float()        tensor.mul_(rgb_range / 255.)        return tensor    return [_np2Tensor(_l) for _l in l]