Tests for the quality of loaded images are not performed

It would be interesting to perform tests for the quality of loaded images. One possibility is the following:

1. An operator verifies that the images being loaded are OK
2. An histogram of target test images are stored in a file
3. When the tests run, the histogram of the new image is compared with the stored histogram, if they match, the test passes.

This could be done for a reduced number of images per database/dataset to save space.

added New Feature label

mentioned in merge request !18 (merged)

This has been done in the update-tests branch.

For raw-loading transforms, the computed histogram is compared to a reference histogram and should match exactly.

For model transforms, the torch.resize does not behave consistently in different environments. The solution was to compute the correlation between the computed histogram and the reference histogram, and allow some small amount of difference.

closed