Input pipeline - reading CSV file for every row

I am woriking on the input pipeline with accelerometer data.
I have a CSV file with following columns, where I have all of the rows:
Using the input pipeline I am trying to read the contents of the file specified in the file_name column (it is a CSV file).
I defined the following map function

def map_ds(row):
    return row[:2],[2])

But the file is read in a binary format, how do I convert it to a tensor when the content looks like this: