将稀疏索引空间转换为形状未知的稠密张量

Question

这个警告是我训练模特时发出的。因为我使用了一个自定义的tensorflow操作，所以培训没有成功，错误是：

2018-08-13 19:56:33.736226: E tensorflow/core/common_runtime/executor.cc:660] Executor failed to create kernel. Not found: No registered '_CopyFromGpuToHost' OpKernel for CPU devices compatible with node swap_out_training/SGD/gradients/pyramid_roi_align_1/CropResize_grad/CropResizeGradBoxes_2 = _CopyFromGpuToHost[T=DT_FLOAT, _class=["loc@traini...radBoxes_2"], _device="/job:localhost/replica:0/task:0/device:CPU:0"](pyramid_roi_align_1/strided_slice/_343)
    .  Registered:  device='GPU'

     [[Node: swap_out_training/SGD/gradients/pyramid_roi_align_1/CropResize_grad/CropResizeGradBoxes_2 = _CopyFromGpuToHost[T=DT_FLOAT, _class=["loc@traini...radBoxes_2"], _device="/job:localhost/replica:0/task:0/device:CPU:0"](pyramid_roi_align_1/strided_slice/_343)]]

我认为这是由我的损失函数中的tf.gather()和{}引起的，因为对于另一个没有这两个函数的模型来说，它可以运行。在

So my question is, am I right? If that so, how to solve this problem?