Automated Ground Truth Generation uses AI tools and machine learning algorithms to produce ground truth data for training machine learning models. This process uses machinebased heuristics and algorithmic computations to significantly reduce the need for human expertise and manual effort. The aim of this project is to create 3D annotations, such as bounding boxes, occupancy grids, and projected segmentation from a video or image sequences. DeepView Enterprise allows for generating ground truth data for more complex areas like 3D as it requires advanced technical skills and the fusion of data from multiple sources. This feature can generate 3D ground truth annotation from a sequence of images / video with minimal with interaction and effort

Multi-functional Precessing Pipeline Stack
Generate 3D Datasets from Raw Videos

Automatically generate segmentation mask in an image
Automatically generate segmentation mask in an image
The segmentation mask error should be less than 15%
Segmentation time per image should be less than 1 second
Be able to track an object over several frames with a tracking error of less that 10% Estimate Tracked object distance with a error or of less than 20%
Be able to process at least 1000 images for tracking
Generate Occupancy Grid with a top-down accuracy per object less than 20%
Occupancy Grid horizontal accuracy of less than 15%
Visualizer to view the generated 3D Ground Truth Annotations with interactions:
Rotate in 3D
Zoom
Pan in 3D
Edit any annotation
Scale in 3D
Move in 3D
View on pre-defined planes
Top-down
Camera normal
Web Based application