Synthetic data could give machines fundamental human cognitive capacities such as object permanence, that has historically been difficult for machine learning models but second nature for humans, according to TRI’s study on multi-object tracking. This new invention improves the resilience of computer vision algorithms, bringing them closer to people’s visual intuition.
At the premier international conference on computer vision, TRI publishes six research papers that push the boundaries of scalable learning.
The Toyota Research Institute (TRI) works to enhance robots, energy and materials, machine learning, and human-centered artificial intelligence through research. TRI’s team of world-class researchers, led by Dr. Gill Pratt, is inventing technologies to enhance human ability, with the goal of making our lives safer and more sustainable. TRI has offices in Los Altos, California, Cambridge, Massachusetts, and Ann Arbor, Michigan, and was founded in 2015.
At the International Conference on Computer Vision, the Toyota Research Institute (TRI) announced the acceptance of six research papers in the field of machine learning (ICCV). The study improves knowledge of a variety of robotic perception tasks, including semantic segmentation, 3D object detection, and multi-object tracking.
Machine learning — the use of computer algorithms that grow with experience and data — has helped TRI researchers make substantial progress in robotics, automated driving, and materials science during the last six years.
“Machine learning is the foundation of our research,” said Dr. Gill Pratt, CEO of TRI. “We are working to create scientific breakthroughs in the discipline of machine learning itself and then apply those breakthroughs to accelerate discoveries in robotics, automated driving, and battery testing and development.”
TRI presented six papers at the International Conference on Computer Vision (ICCV), exhibiting TRI’s robust machine learning research in areas such as geometric deep learning for 3D vision, self-supervised learning, and simulation to real or “sim-to-real” transfer.
“Within the field of machine learning, scalable supervision is our focus,” said Adrien Gaidon, head of TRI’s Machine Learning team. “It is impossible to manually label everything you need at Toyota’s scale, yet this is the state-of-the-art approach, especially for Deep Learning and Computer Vision. Thankfully, we can leverage Toyota’s domain expertise in vehicles, robots or batteries to invent alternative forms of scalable supervision, whether via simulation or self-supervised learning from raw data. This approach can boost performance in a wide array of tasks important for automated cars to be safer everywhere anytime, robots to learn faster and battery development to speed up lengthy testing cycles.”
In addition, the results of TRI’s pseudo-lidar study reveal that large-scale self-supervised pre-training improves the performance of image-based 3D object detectors significantly. The suggested geometric pre-training allows powerful 3D Deep Learning models to be trained using limited 3D labels, which are expensive or impossible to get from photos alone.
For more information about TRI, please visit http://tri.global.