RealSim enables perception engineers to obtain more variety in their data with fewer real-world annotated images. This reduces cost without compromising model performance.
“We’ve seen models train comparatively using RealSim data as they would using annotated images logged from the real world.”
Prof. Devdatt Dubhashi – Chief Scientist
USE CASE
We’re developing RealSim in automotive/AD, off-road domains such as mining, and even manufacturing.
RealSim has a substantial amount of pre-trained knowledge. It only requires a limited volume of real-world images to fine-tune to a new domain.