The current test suite generates a human readable report from the attacks and json result files.
But it doesn't provide a list of result/image files that failed and another list of file names where detection succeeded.
These lists could be useful for pruning the test image set or for generating other stats from the result files.
I.e. the current report generation should be extended to also provide two files listing the names of cases that failed and succeeded with one file name per line, so scripts can use these as input for further processing.