On-Line Help for Web Based Verification of Concrete Image Segmentationn

Concrete Sample Segmentation:

The inspection of 2D concrete segmentation into damage sub-classes is enabled by presenting triplets of raw fields of view (FOVs), color-coded masks with damage sub-classes, and color coded masks with contextual labels used to derive damage labels.
The manually assigned verification labels for each damage segmentation span good, correct, bad colors, and bad pixels.

• Good:Visually, the colors assigned to pixel locations are accurate (at least 90%), this segmentation is valid.
• Correct: Visually, the colors assigned to pixel locations accurate (at least 75%), but some small parts can be over or under segmented, this segmentation is valid.
• Bad colors: The colors assigned to correct pixel locations are not accurate enough (less than 75%), this segmentation should be discarded.
• Bad pixels: The pixel locations are not accurate enough (less than 75%), this segmentation should be discarded.

One can step through all triplets of concrete segmentation results by clicking on the buttons "Next" or "Previous". Verification labels are assigned to each segmented FOV by clicking on one of the buttons with the labels under the heading "How would you classify this semantic segmentation?". A user can see already assigned or updated labels by clicking on "Display CSV below". When a user completes the labeling, all assigned labels can be downloaded by clicking on the button "Download CSV".
WARNING: Please, do not reload current web pages during labeling since you would lose all already assigned labels. All labels are kept in the browser cache during a labeling session and therefore switching to a different tab or web page will cause losing the labels.