Estimation for Each Type of Annotation
| SN | Type of annotation and labeling | Mean Annotation time for Lower complexity for single image / Frame | Mean Annotation time for Higher complexity for single image / Frame |
|---|---|---|---|
| 01 | Instance based Semantic segmentation for camera | 13 minutes | 46 minutes |
| 02 | 2D bounding box from the camera | 0.91 minutes | 2.3 minutes |
| 03 | 3D bounding box from the camera | 2.75 minutes | 4.5 minutes |
| 04 | Global scene labeling | 0.39 minutes | 1.3 minutes |
| 05 | 3D bounding box in point cloud | 4 minuts | 8.5 minutes | 06 | Point annotation for point cloud | 15.6 minuts | 40 minutes |
Estimation Factors for Annotation & Labeling
01.
Inputs from Customer
• Bounding Box Type • Bounding Box Tightness • Input Data Format • Unique Instance ID needed • Label File Format • Multiple class labels for same pixel • Billing Unit (Images) • Required Spec available? • Bound Box Rotation needed? • Object of Interest
02.
Effort Estimation (Manual)
Annotation
Labelling Effort
Time to Process 1 frame
Seconds
Time to assign Labels (Unique
instance ID + multiple class
labels)
Seconds
03.
Effort Estimation (Manual)
Verification
Effort
Verification of 1 frame
Seconds
Verification of labels
Seconds
Total time per object
Seconds
04.
Effort Estimation (Manual)
Next image annotation
Effort
Time to move next image
Seconds
Analysis of image
Seconds
Analysis of image
Seconds
05.
Other factors
• Semi-Auto mation if image sequence
• Tracker level automation
(Semi Automation)
*Only possible in case of continuous
data
• Automation from Machine Learning for
object detection (Bootstrap method)