当你load tensorflow 的meta文件时new_saver tf.compat.v1.train.import_meta_graph(path to meta)出现如下的错误Op: MatMul Node attrs: transpose_bfalse, TDT_INT32, transpose_afalse Registered kernels: deviceXLA_GPU; T in [DT_FLOAT, DT_DOUBLE, DT_COMPLEX64, DT_BFLOAT16, DT_HALF] deviceXLA_CPU; T in [DT_FLOAT, DT_DOUBLE, DT_COMPLEX64, DT_BFLOAT16, DT_COMPLEX128, DT_HALF] deviceXLA_CPU_JIT; T in [DT_FLOAT, DT_DOUBLE, DT_COMPLEX64, DT_BFLOAT16, DT_COMPLEX128, DT_HALF] deviceXLA_GPU_JIT; T in [DT_FLOAT, DT_DOUBLE, DT_COMPLEX64, DT_BFLOAT16, DT_HALF] deviceGPU; labelcublas; T in [DT_HALF] deviceGPU; labelcublas; T in [DT_COMPLEX128] deviceGPU; labelcublas; T in [DT_COMPLEX64] deviceGPU; labelcublas; T in [DT_DOUBLE] deviceGPU; labelcublas; T in [DT_FLOAT] deviceGPU; T in [DT_HALF] deviceGPU; T in [DT_COMPLEX128] deviceGPU; T in [DT_COMPLEX64] deviceGPU; T in [DT_DOUBLE] deviceGPU; T in [DT_FLOAT] deviceCPU; labeleigen; T in [DT_COMPLEX128] deviceCPU; labeleigen; T in [DT_COMPLEX64] deviceCPU; labeleigen; T in [DT_INT64] deviceCPU; labeleigen; T in [DT_INT32] deviceCPU; labeleigen; T in [DT_BFLOAT16] deviceCPU; labeleigen; T in [DT_HALF] deviceCPU; labeleigen; T in [DT_DOUBLE] deviceCPU; labeleigen; T in [DT_FLOAT] deviceCPU; T in [DT_COMPLEX128] deviceCPU; T in [DT_COMPLEX64] deviceCPU; T in [DT_INT64] deviceCPU; T in [DT_INT32] deviceCPU; T in [DT_BFLOAT16] deviceCPU; T in [DT_HALF] deviceCPU; T in [DT_DOUBLE] deviceCPU; T in [DT_FLOAT]# 把clear_device设成True即可解决new_savertf.compat.v1.train.import_meta_graph(path to meta,clear_devicesTrue)问题解决
和tf.get_collection()的用法)
 Net Surgery (卷积，全连接的weights和biases的参数存储方式理解))
