华为计算微信公众号
昇腾AI开发者公众号
华为计算微博
华为计算今日头条
今天初始化Tensor时发现一个问题，使用下面第一种Shape的顺序：B*C*H*W和第二种顺序B*H*W*3打印出来的Shape是一样的，这个初始化对Shape的顺序有什么要求吗
# B*C*H*W std::vector<unsigned int> batchShape = { static_cast<unsigned int>(batchSize), 3, static_cast<unsigned int>(imageHeight), static_cast<unsigned int>(imageWidth) }; Tensor batchTensor(batchShape, TensorDType::FLOAT32, -1); # p batchTensor.GetShape()[0] # $1 = (__gnu_cxx::__alloc_traits<std::allocator<unsigned int>, unsigned int>::value_type &) @0xaaaaadb6b4e0: 5 # (gdb) p batchTensor.GetShape()[1] # $2 = (__gnu_cxx::__alloc_traits<std::allocator<unsigned int>, unsigned int>::value_type &) @0xaaaaad8fc144: 512 # (gdb) p batchTensor.GetShape()[2] # $3 = (__gnu_cxx::__alloc_traits<std::allocator<unsigned int>, unsigned int>::value_type &) @0xaaaaadb6b648: 640 # (gdb) p batchTensor.GetShape()[3] # $4 = (__gnu_cxx::__alloc_traits<std::allocator<unsigned int>, unsigned int>::value_type &) @0xaaaaadb6b4ac: 3 # B*H*W*C std::vector<unsigned int> batchShape = { static_cast<unsigned int>(batchSize), static_cast<unsigned int>(imageHeight), static_cast<unsigned int>(imageWidth), 3 }; Tensor batchTensor(batchShape, TensorDType::FLOAT32, -1); # p batchTensor.GetShape()[0] # $1 = (__gnu_cxx::__alloc_traits<std::allocator<unsigned int>, unsigned int>::value_type &) @0xaaaaadb6b670: 5 # (gdb) p batchTensor.GetShape()[1] # $2 = (__gnu_cxx::__alloc_traits<std::allocator<unsigned int>, unsigned int>::value_type &) @0xaaaaad8fc0e4: 512 # (gdb) p batchTensor.GetShape()[2] # $3 = (__gnu_cxx::__alloc_traits<std::allocator<unsigned int>, unsigned int>::value_type &) @0xaaaaadb6bad8: 640 # (gdb) p batchTensor.GetShape()[3] # $4 = (__gnu_cxx::__alloc_traits<std::allocator<unsigned int>, unsigned int>::value_type &) @0xaaaaadb6b11c: 3
我要发帖子
点赞
收藏
回复
分享
举报
今天初始化Tensor时发现一个问题，使用下面第一种Shape的顺序：B*C*H*W和第二种顺序B*H*W*3打印出来的Shape是一样的，这个初始化对Shape的顺序有什么要求吗
# B*C*H*W std::vector<unsigned int> batchShape = { static_cast<unsigned int>(batchSize), 3, static_cast<unsigned int>(imageHeight), static_cast<unsigned int>(imageWidth) }; Tensor batchTensor(batchShape, TensorDType::FLOAT32, -1); # p batchTensor.GetShape()[0] # $1 = (__gnu_cxx::__alloc_traits<std::allocator<unsigned int>, unsigned int>::value_type &) @0xaaaaadb6b4e0: 5 # (gdb) p batchTensor.GetShape()[1] # $2 = (__gnu_cxx::__alloc_traits<std::allocator<unsigned int>, unsigned int>::value_type &) @0xaaaaad8fc144: 512 # (gdb) p batchTensor.GetShape()[2] # $3 = (__gnu_cxx::__alloc_traits<std::allocator<unsigned int>, unsigned int>::value_type &) @0xaaaaadb6b648: 640 # (gdb) p batchTensor.GetShape()[3] # $4 = (__gnu_cxx::__alloc_traits<std::allocator<unsigned int>, unsigned int>::value_type &) @0xaaaaadb6b4ac: 3 # B*H*W*C std::vector<unsigned int> batchShape = { static_cast<unsigned int>(batchSize), static_cast<unsigned int>(imageHeight), static_cast<unsigned int>(imageWidth), 3 }; Tensor batchTensor(batchShape, TensorDType::FLOAT32, -1); # p batchTensor.GetShape()[0] # $1 = (__gnu_cxx::__alloc_traits<std::allocator<unsigned int>, unsigned int>::value_type &) @0xaaaaadb6b670: 5 # (gdb) p batchTensor.GetShape()[1] # $2 = (__gnu_cxx::__alloc_traits<std::allocator<unsigned int>, unsigned int>::value_type &) @0xaaaaad8fc0e4: 512 # (gdb) p batchTensor.GetShape()[2] # $3 = (__gnu_cxx::__alloc_traits<std::allocator<unsigned int>, unsigned int>::value_type &) @0xaaaaadb6bad8: 640 # (gdb) p batchTensor.GetShape()[3] # $4 = (__gnu_cxx::__alloc_traits<std::allocator<unsigned int>, unsigned int>::value_type &) @0xaaaaadb6b11c: 3