Synthetic Data
for Computer Vision Model Training
Why Synthetic Data?
What is often a large obstacle for ML/Deep learning applications is the acquiring of good data to better train its algorithms. There are many reasons as to why this is difficult:
Data can be:
-expensive
-non-existent
-hindered by privacy concerns or legal issues
-incorrectly or not fully labeled (human error is often to blame)
Domains which Synthetically is currently working on:
What Synthetically offers:
•3d room environments with
•rgb images
•normal projections
•depth maps
•precise semantic segmentation
•floors / walls / ceiling
•furniture / rugs
•exact camera placement and configuration in an environment
•deltailed and varied environments, allowing algorithms to better generalize better generalize
•customizable parameters
•lighting
•varied room geometries
•furniture
•floor and wall “materials”
•concrete
•carpet
•wood floors
•lots of data
Use Cases:
Robotics
Augmented Reality
Diminished Reality
Indoor Semantic Segmentation