Instance Segmentation

Try to use Instance Segmentation.

Among the segmentation in the form of Polygon Labeling, it is a smart labeling model that performs a distinguishable Instance Segmentation task for each instance within the same class.

It is available when the task type is set to Poly segmentation in the project's labeling setting.

Example of Instance Segmentation Smart Labeling

Vehicle/Pedestrian

1. Definition of Class

• Vehicle
  Standard vehicles such as Sedan or SUVs (except vans, trucks, buses, and special vehicles, excluding vehicle accessories)
• Pedestrian
  Kids, adult pedestrians, and riders

2. Minimum inference size

• Vehicle : 20px(width) * 25px(height)
• Pedestrian : 10px(width) * 15px(height)

---

Lane/Corner

1. Definition of Class

• Lane: Lanes on the road (colors of lanes: white, orange, blue), (types of lanes: Solid line, dotted line)
• Corner: Curbs at the edge of the road

2. Minimum inference size

• Line : px(width) * px(height)
• Corner : px(width) * px(height)

---

The extent of the performance guarantee of Smart Labeling

1. Information about images

• Resolution: HD or higher
• Camera lens view angle: 120° or lessHowever, inference performance will be degraded for images that have not been corrected for distortion. e.g., fisheye lens

2. Threshold

• Performance will be degraded for inference instances that are less than threshold 0.8.

3. Exceptions

• Inference performance will be degraded for images that blur occurs in images because of rain, snow, foreign objects, etc.
• Inference performance will be degraded for images that have severe backlighting.

---

If you have any other inquiries, please get in touch with us at [email protected]