This article describes how to use the bounding box tool to make box annotations as a contributor.
Note: Contributors should always rely on the individual job instructions above anything mentioned in this article. For example, unique tool features or a requested way to place the box borders vs. the target object may vary. Where they are different, always consider the job's instructions to be the true requirements of the task.
Hotkeys and Cursor Actions
Place your cursor over a button to see text stating what it does, as well as an associated 'hotkey' combination you can use to invoke that action via keyboard.
When images in the tool exceed the size of your browser viewport, you will see a red bar on the borders where any obscured portion of the image (green borders show that the image is completely displaying up to that particular edge). Cut-off images can be moved around via scroll actions when your cursor is over the image.
Click on inside a box to move it around or resize -- it will turn blue (inactive boxes are yellow) when you have done so.
Click on the dot outside of the bounding box -- you will be able to rotate the box.
Adding a Box:
|
 |
|
 |
Removing a Box:
|
 |
Nothing to Box:
|
 |
Zoom:
|
 |
Enhance:
Clicking the star icon, will enhance the image increasing the contrast so objects may be easier to identify
Undo/Redo:
Focus:
When the focus tool is enabled, only the selected box will be shown. This is helpful when there are many objects to be labeled in the image.
Hide:
Hide will hide all boxes drawn on the image
Popout:
- Opens the tool in a new full screen window
- To save the image, click on the icon in the top right corner. This will close the pop-out, but save your boxes to the original task page allowing you to submit
Prediction Slider
In some cases, you may see a 'slider' control on the top bar and callouts next to pre-labeled boxes that have numeric percentages in them.
These jobs offer these pre-guessed boxes and associated confidence scores of those boxes as a starting point to potentially save time. If the boxes you see are too numerous or inaccurate, click-drag the slider to the right to drop the lower-confidence boxes out. If there are many unlabeled objects, click-drag the slider left to include more, lower-confidence boxes. When you have chosen an optimum starting point, you can delete and correct any boxes as needed as well as add others from scratch. When you attempt to submit, all boxes that are displaying will be equally considered as submissions regardless of whether they were left as-is, changed or created newly. Further questions about this feature should be directed to your Customer Success Manager.
Test Question Review
When the provided answer to a test question is wrong, you will see an image like the following with red and green boxes.
- Red boxes are the incorrect annotations
- Green boxes are the correct annotations.
- If you submit a judgment saying that there is ‘Nothing to Box’ but there are objects to box, you will see the message "You did not annotate the image but the test question is annotated. The test question annotation is below." and the expected annotations in Green
- If you submit a judgment that has boxes, but the author did not have any boxes, you will see the message "You annotated the image but the test question has no annotations. Your annotation is below." and your annotations in Red
- Note: Unlike most other Test Question 'misses', written Reason content explaining correct box positions is not shown for Bounding Box questions, as the boxes are self-explanatory of the Task Author's expectations.
