Flagging images for testing
Images can be flagged for testing using context menu in Images list and the Flag images as test command:
Selected images will be assigned a test flag which is reflected in their green color:
TIP: Ctrl+T keyboard shortcut performs the same action.
Note, that the state of the selected images changes by applying this command. Therefore, images that were already part of the test set will become part of the training set (will loose the green color emphasis).
Additional image flagging commands
Data menu contains several additional commands for image flagging in Image flags sub-menu:
- Selected images may be all flag for training or for testing
- Percentage of of selected images can be randomly assigned a test flag. This is useful to perform manual cross-validation on a large number of scans. We select all scans (Ctrl+A) and flag random fraction as test. We rebuild a model and record test set performance. Note, that in order to retrain a model, we need to have labeled data. Therefore, most of your scans should be meaningfuly annotated in order to use this manual cross-validation approach.