protobuf/init.lua:669: bad argument #1 to 'ReadTag' (string expected, got userdata)

[arashno]

Hi all,
I have an LMDB which is created in python.
Each record is a protobuffer message.
I am trying to decode the messages by doing this:

key,value=cursor:get() local record = Record_pb.Record() record:ParseFromString(value)

The 'value' is a Tensor of Bytes.
and it gives me the following error. What should I do?

.../install/bin/luajit: .../torch/install/share/lua/5.1/protobuf/init.lua:669: bad argument #1 to 'ReadTag' (string expected, got userdata)
stack traceback:
[C]: in function 'ReadTag'
.../torch/install/share/lua/5.1/protobuf/init.lua:669: in function '_internal_parse'
.../torch/install/share/lua/5.1/protobuf/init.lua:687: in function 'merge_from_string'
.../torch/install/share/lua/5.1/protobuf/init.lua:696: in function 'ParseFromString'
LuaRead.lua:17: in main chunk
[C]: in function 'dofile'
...ouzz/torch/install/lib/luarocks/rocks/trepl/scm-1/bin/th:145: in main chunk
[C]: at 0x00405800

[djungelorm]

Looks like you are passing a value of type userdata to ParseFromString but it expects a Lua string. I guess you need to convert this userdata value to a string first? Not sure what you mean by a tensor of bytes (I've no experience with LMDB)

[arashno]

That's true.
By a tensor of bytes I mean a Torch's ByteTensor.
I read a tensor of bytes from DB (an array of bytes), Is there any fast tool to convert it to a string?
Thanks

[djungelorm]

There's nothing built into Lua unfortunately. I think the quickest way would be something like:

function FastBytesToString(bytes)
  s = {}
  for i = 1, getn(bytes) do
    s[i] = strchar(bytes[i])
  end
  return concat(s)
end

Found it on here: http://lua-users.org/lists/lua-l/2002-03/msg00166.html

[arashno]

I tried that, it works but it makes my program very slow.
Now my problem is a little bit different.
Now I can read a string from the DB.

This is my prototxt file:

message Record{

optional int32 channels = 1;
optional int32 height = 2;
optional int32 width = 3;
// the actual image data, in bytes
required bytes data = 4;
required int32 label = 5;
optional int32 count = 6;
optional float standing = 7;
optional float resting = 8;
optional float moving = 9;
optional float eating = 10;
optional float interacting = 11;
optional float babies = 12;
}

The problem is about 'data' field.
I read a string, call the ParseFromString method on it and I expect a tensor of bytes as the 'data' field but it gives me another string.
I have to convert back the string into a tensor of bytes and it makes my program very slow.