Same Image Twice
Testing the same image again to verify the results are the same.
Generating Results
This test will use the pipeline in the same way as the previous test using the same image as before. The job ID of the previous test is 874594801b. The second job has ID c256255c68.
Comparing Results
Validity of the Calculation
Before the results themselves are compared, it is important to see if the calculation itself is correct. The goal of the analysis module is to determine whether the given AU value is within the maximum deviation of the reference value. To calculate this, the minimum and maximum values are calculated, after which it is determined whether the given value is between these two. The code can be seen here.
Taking an example from job 874594801b, the first AU is True and the other False.
Time to manually calculate this to see if these results are correct, starting with AU01_c. The relevant values for it are as follows (taken from the associated parameter set and action unit report):
Measured AU value: 0
Reference value: 0
Maximum deviation: 5
As the measured value and reference value are equal, this is clearly correct. As for AU01_r:
Measured AU value: 0.68
Reference value: 1
Maximum deviation: 6 - meaning the measured value would have to be between 0.94 and 1.06
As can be seen, 0.68 is not between 0.94 and 1.06, and therefore not within reference parameters. Because of this, it is marked as False.
Comparing the Results
First, it is important to look at the results of the analysis performed earlier in this article and the analysis mentioned previously.
As can be seen, the results of the analysis are the same. As an extra precaution, the core OpenFace analysis data will also be compared. As there is no dedicated page to view this, it will be done using a database query.
Conclusion
Analysing the same image twice provides the same results, indicating the pipeline is internally consistent.
Last updated
Was this helpful?