During the prediction, IKOSA AI applications will assign an ID to each found object listed in the analysis job results. These IDs not only determine the order in which the instances are listed in the CSV or XLSX output, but they can also help find the instance within the visualization.
This page explains the unique aspects of the way instance numbering works in the visualizations.
Order of the instances on small images
The instances on small images (smaller than a “tile”, more on this under the next point) are ordered by a line-dominant scheme. This means the image is read line by line and from left to right when numbering the instances. The instance’s relevant point is the upper left corner of its bounding box. Simplified, this can be imagined like this:
Instance order on large images
Large images are split within the application into tiles (dashed line on the image below). The numbering on each of those tiles follows a line-dominant pattern (described in the previous section). However, neighboring tiles are numbered in sequence from left to right and top to bottom. This can result in a pattern like the one below.
The default tile size for IKOSA AI Applications is 2048 x 2048 Pixels. So, if your image is more significant than 2048 Pixels in any direction, expect your instance numbering to follow this pattern.
Numbers are omitted on small instances
When labeling small instances, at some point, their instance ID hides them behind its index visualization. To avoid this, a visualization threshold for the instance number exists. Numbers will not be displayed if the bounding box is over 5 times larger than the instance area.
Limit the number of instances to be numbered at all
When predicting instances on large images or whole slide images (WSI), the predicted number can exceed what can be reasonably analyzed visually. There is a maximum number of instances until which the instance numbers for all of them will be displayed. The reason for this is that instances with high indices would require a large size to display their value.
By default, this number threshold is 100.000 instances. So, if your prediction contains more than that, no instance number will be shown in the visualization.
If you still have questions regarding your application training, feel free to send us an email at support@ikosa.ai. Copy-paste your training ID in the subject line of your email.