Skip to content

How to interpret the result files

Anh Nguyen edited this page Jan 24, 2015 · 1 revision

The results of an experiment is stored in the mmm folder generated. In this folder, you will find:

  1. Archive files.
  • E.g. archive_1000.dat, which is the file generated after 1000 generations containing the fitness or prediction scores of all 1000 images generated.

  • You will find two columns of numbers, which look something like below:

     7 0.0727381
     8 0.0317073
     9 0.0296667
     10 0.0185064
    
  • 10 here is the category #10 in 1000 ImageNet categories. (This column ranges from 0 to 999)

  • 0.0185064 is the prediction score of best image evolved in that category. In this case, the image has the confidence of ~1.85%.

  1. Images folders.
  • E.g. map_gen_1000, which is the folder generated after 1000 generations containing the final images evolved. These images correspond to the entries in the archive_1000.dat.
  • A file would look like: map_1000_9_0.32.png. This means that:
    • This file is generated after 1000 generations
    • This image is the best (highest score) in the category #9 out of 1000 ImageNet categories.
    • This image is given 32% prediction score by the DNN model.
Clone this wiki locally