Max Robotics — Global Service Robot Directory + US Certification

About

Song Bai is a researcher whose work sits at the intersection of computer vision, autonomous driving, and 3D scene understanding. His most impactful contribution, "Importance-Aware Semantic Segmentation in Self-Driving with Discrete Wasserstein Training" (2020, 54 citations), tackles a critical challenge in perception for self-driving cars. Bai moved beyond standard cross-entropy loss by introducing a discrete Wasserstein training framework that accounts for the varying importance of different semantic classes, significantly improving mean Intersection-over-Union (mIoU) performance—a key metric for safe navigation. This work directly addresses the real-world need for robust, pixel-level classification in dynamic environments. Additionally, Bai has pioneered the emerging field of 2D image-based 3D scene retrieval (2018), a novel paradigm that allows users to search for relevant 3D scenes using a simple 2D photograph. This intuitive framework bridges the gap between 2D perception and 3D databases, with applications in robotics, augmented reality, and content-based retrieval. By advancing both semantic segmentation for autonomous systems and cross-modal 3D search, Bai demonstrates a clear trajectory toward making machines perceive and interact with the world more intelligently.