[QUESTION]Does the tensor created by nvcv::TensorWrapData take the ownship of external memory?

[lazylazypig]

Does the tensor created by nvcv::TensorWrapData take the ownship of external memory?
the codes 'samples\classification\Main.cpp' make me confused:

// Allocating memory for input image batch
nvcv::TensorDataStridedCuda::Buffer inBuf;
...
CHECK_CUDA_ERROR(cudaMallocAsync(&inBuf.basePtr, batchSize * inBuf.strides[0], stream));

nvcv::Tensor::Requirements inReqs
    = nvcv::Tensor::CalcRequirements(batchSize, {maxImageWidth, maxImageHeight}, nvcv::FMT_RGB8);

nvcv::TensorDataStridedCuda inData(nvcv::TensorShape{inReqs.shape, inReqs.rank, inReqs.layout},
                                   nvcv::DataType{inReqs.dtype}, inBuf);

nvcv::Tensor inTensor = TensorWrapData(inData);
...

as shown above, we use cudaMallocAsync to malloc external memory and bind to inBuf.basePtr. but finally, we can't see any cudaFree/cudaFreeAsync to free the allocated memory.

1. so, does it mean that the tensor created by TensorWrapData(...) takes the ownship of the external allocated memory? or it was just a little mistake that you forgot to call cudaFree/cudaFreeAsync?
2. if it's real that tensor takes the ownship of the external allocated memory, what can i do to avoid that? in other words, how can i wrap the external allocated memory to a tensor without lost the ownship?

[sergeybayandin]

Does specify cleanup callback (TensorDataCleanupCallback) make any differences?