TVMean produces different mean image outputs depending on whether full paths are used in the input text file. I tested this the following ways:

1. Standard, no full paths. Running in a single directory, with all input files in the current directory. Text input file contains only the names of the input files.
2. Running in a single directory, all input files in the current directory, but the input text file contains full paths to those files. This produces results different from 1.
3. Running in a directory different from where the input files are located. Full paths to those files are included in the text input file. This produces results identical to 2.

This turns out to be a failure in the input text file parsing, where if there is no newline character after the last input file, it does not read the last subject. It just so happened that the input files I tested this with differed in this regard.