top of page
Search
Randomize Everything. Just Don't Mix the Domains.
Domain randomization is supposed to stop a model from memorizing its surroundings. Our thermal-imaging experiment showed us exactly where that idea stops working. Synthetic data comes with a familiar rule: if you don't want the model to memorize something, randomize it. Change the lighting. Change the camera angle. Change the texture. Swap the background, again and again. Do it enough, and the model should stop caring about the scenery and start learning the object itself. Th
Finn Chen
13 hours ago4 min read
Not Just Relacing Real Data with Synthetic. Multiply It.
Synthetic data can sure replace real data in most cases, but adding a bit of real date can help calibrate and anchor the models for better results.
Finn Chen
Sep 45 min read
Innovation Endorsed: Landmark Patent for Physical AI Granted by U.S. Patent and Trademark Office (USPTO)
"Machine-learning Method on Vectorized Three-Dimensional Model and Learning System Thereof" describes a novel method for utilizing vectorized 3D models to generate synthetic data of real-world objects for AI training. It can be applied for various applications, including object identification and anomaly detection, where training data are difficult or economically infeasible to acquire, and the robotic agents need these capabilities to function and swiftly interact with the p
Ting-Yuan Wang
Jan 52 min read
bottom of page