Image labeling tool

SentiSight.ai offers a powerful image labeling tool that allows you to draw bounding boxes, polygons, bitmaps, polylines and points. We call all of these labels "object labels". Images that are labeled by these labels are marked by this symbol.

The object labels are used for object detection (available on Sentisight.ai) and segmentation (currenly available as custom project only) model training. Additionally, object labels can be downloaded as .json and used for your in-house model training.
Object labels made count towards "Labels made" statistic and appear in the list of project's labels. "Synchronise labels" setting also makes it so object labels added also label the image for classification with same label. Below you can see basic types of object labels you can use in the labeling tool.
Bounding boxes are simple rectangular shapes. You can set fixed aspect ratio for bounding boxes in the settings. ”Image
Polygons are complex shapes defined by multiple points connected by lines. A single object can consist of multiple polygons and/or holes. ”Image
Polylines are collections of points connected by lines, useful for labeling things which are defined by structure, not precise shape. ”Image
Points are useful for labeling miniscule objects. ”Image
Bitmaps are freeform hand-drawn masks that can be used to label objects that have complex shapes. ”Image

Labeling tips

While points and bounding boxes are very simple to create and edit, there is a number of tricks to other object types.

You can click on images to open larger animated gifs.

lt-pg-pic

Polygons

When drawing a polygon, press Enter or New object (shortcut N) to finish drawing a current polygon. Next click will start a new object.

Select the polygon and press Edit (shortcut E) to add or move vertices. To move a vertex just click on it and drag. To add new vertices, click close to a border - you will see a projection of new vertex and changed borders when in position to add a point.

Press New polygon in group (shortcut G) to create a new separate polygon that will be considered part of the same object - useful for labeling multipart objects.

Press New hole (shortcut H) to create a hole. Holes are created and edited exactly like polygons, but parts of the image covered by them are not considered part of the object the hole belongs to.

lt-pl-pic

Polylines

When drawing a polyline, press Enter or New object (shortcut N) to finish drawing current polyline. Next click will start a new object.

Select the polyline, and press Edit (shortcut E) to move points.

Press Add points (shortcut K) to add points - new point will be linked to the previous created point. There is currently no functionality to add points in the middle of the polyline.

Press Remove points (shortcut R) and click on individual points to remove them. The polyline will redraw itself with remaining points - deleting a point does not create a gap.

lt-bm-pic

Bitmap

Bitmaps are drawn with a MS Paint-like brush. Unlike other objects, bitmaps do not have to be connected in any way to form a single object. You paint any number of disjointed masks in a single object.

When drawing a bitmap, press Enter or New object (shortcut N) to finish drawing a current bitmap. Next click will start a new object.

Select the bitmap and press Draw (shortcut E) to add to existing bitmap.

Select a bitmap and press Erase (shortcut R) to replace the brush with an eraser that will remove mask.

Press Fill (shortcut G) and click inside a closed bitmap shape to fill it forming a solid mask. Note that if bitmap area is not closed entire image will be filled.

Shared features

The following features are applicable to most object types.
This icon will mark object as "occluded". It will change the object's border to a dotted line to show that it is partially obscured by other objects. Currently, this is purely for information purposes and has no effect on model training.

Visibility allows you to hide the object label, making it invisible and non-interactive, allowing you to better see or edit objects in same area. Alternatively you can use "Hide all objects" and "Show all objects" buttons to change visibility of all objects present in an image.

Attributes are text-only notes that can be added to objects.

Keypoints

Keypoints can be added to bounding boxes and polygons, marking position of important points - e.g. eyes and nose on a face.
You can save a template for a set of keypoints with certain names for a specific object label. For instance, if you add keypoints named "Left eye", "Right eye", "Nose", etc to a bounding box labeled "Face" and save the template, next time you add keypoints to an object labeled "Face" first keypoint will be automatically named "Left eye", the second one "Right eye" and so on.

Rasterization

Bounding boxes and polygons can be converted to bitmap masks using Rasterize button. Additionally Rasterize All Objects button is present to turn all applicable objects into bitmaps.

Please note that rasterization clears all keypoints and attributes from the object.


Converting bitmap to polygon

Additionally, bitmaps can be converted to polygons using Convert to polygon button.

You can choose to convert a bitmap to very detailed polygon perfectly resembling it or use "Simplify polygons converted from bitmap" setting to create a much simpler polygon.


Smart labeling tool

Smart labeling tool is a powerful instrument that allows you to quickly create complex bitmap masks by separating objects from the background.

All it needs is a user to mark some points of the foreground and background, from which the algorithm extrapolates the rest.

Using smart selection tool

1. Select a rectangular area that the tool will work in.

2. Mark the foreground. For largely homogenous looking regions that contrast well with the background a few brief lines is enough. For regions with complex colors that may blend with the background it might be better

3. Mark the background. In general you don't have to be very precise.

4. Press Extract to have the tool mask the foreground. Examine it carefully to make sure it covers all parts of the image you want.

5. Repeat steps 2-4 until you are satisfied with the result.

6. Once you are content with the result, press Done to produce a bitmap mask. You can touch it up with Bitmap tool if you need.

Please note that until you press Done the work done with smart selection tool cannot be saved and will be lost if you click on another tool, switch image, or close labeling tool.

You can see workflow of smart selection tool in the gif below.