Has Alibaba, which recently established the “Dharma House,†also entered the world of autonomous driving?
On January 14th, a reporter from Ayao discovered that Alibaba made its debut on the KITTI road scene segmentation chart, one of the most prestigious benchmarks for computer vision in autonomous driving.
The ranking shows that Alibaba’s iDST team secured first place in three segmentation tasks, including two specific scenarios—UU_ROAD (Urban Lane) and UMM_ROAD (Multi-lane), as well as overall performance for URBAN_ROAD (urban lane).
Alibaba confirmed the news to Xinhua News but clarified that it is not entering the autonomous driving business directly. Instead, the company emphasized its focus on fundamental scientific research, with autonomous driving technology being one of its key areas of interest. The research is led by Ren Xiaofeng, an Alibaba machine vision scientist.
The KITTI dataset is widely used as a benchmark for autonomous driving development, with companies like Toyota, Samsung, and NEC also appearing on the rankings.
Founded by the Karlsruhe Institute of Technology in Germany and the Toyota Institute of Technology in the U.S., the KITTI dataset is recognized globally as one of the largest and most comprehensive datasets for evaluating autonomous driving algorithms. It covers five main scenarios: urban areas, residential zones, roads, campuses, and pedestrian environments.
The dataset includes bird's-eye view perspectives, making it highly relevant for autonomous systems. It features complex road structures, such as hard-to-detect lanes and edge regions, which are critical challenges in real-world applications.
According to the KITTI website, the evaluation dataset is divided into categories such as scene classification, object detection, and tracking. A researcher told Xinhua News that road scene segmentation plays a vital role in autonomous driving, supporting functions like driving area division, path planning, high-precision mapping, and AR navigation.
Ali won first place in three segmentation tasks. Source: KITTI official website screenshot
While many autonomous driving teams can handle large-scale urban road segmentation, accurately identifying road edges—such as stones, curbs, or parked cars—remains a challenge. Improving edge segmentation accuracy is a major focus of current research.
Alibaba’s iDST team, led by Ren Xiaofeng, introduced an online learning method based on difficult sample mining. They also added an online data augmentation module, along with mechanisms for integrating global and local features. These enhancements improved the model’s ability to address issues like inaccurate edge segmentation, shadow interference, and surrounding road recognition.
After six months of research, Alibaba achieved a segmentation accuracy of 96.06% for U.S. lanes and 97.70% for multi-lane scenarios. It also secured first place in the overall URBAN_ROAD evaluation with a 96.76% accuracy rate—marking its first appearance on the KITTI road scene segmentation chart.
In June 2017, Handan News reported that Ren Xiaofeng, a top Chinese scientist at iDST, had previously led Amazon Go’s unmanned store algorithm. Upon joining Alibaba, he was praised by experts as rare in the field of computer vision: “He has a comprehensive and deep understanding of image classification, object recognition, tracking, detection, and event detection.â€
The reporter also found that Alibaba’s iDST team appeared in several other KITTI projects, such as vehicle detection, pedestrian detection, and re-identification. Some of these technologies have already been deployed in Alibaba Cloud’s ET City Brain, currently used for urban traffic management.
Alibaba is continuously strengthening its scientific capabilities and investing heavily in basic research. At the 2017 Yunqi Conference, the company announced the establishment of the “Dharma House†lab to explore future science and technology. It plans to invest over 10 billion yuan in new technologies over three years, focusing on foundational research and disruptive innovation. Jack Ma, Alibaba’s chairman, hopes the Dharma House will become one of the best research institutions in the world.
Contact Switch,Reed Contact Switch,Magnetic Contact,Magnetic Contact Switch
Shanghai Janetec Electric Co., Ltd. , https://www.janetecelectric.com