About

Zhen Lei is a professor of energy and environmental economics at Penn State's John and Willie Leone Family Department of Energy and Mineral Engineering. He holds a Ph.D. in Agricultural and Resource Economics from the University of California at Berkeley and a Ph.D. in Chemistry and Pharmaceutical Sciences from Peking University Health Sciences Center. His expertise spans energy and environmental economics, health economics, science and technology policy, economics of innovation and intellectual property, big data and data science, industrial organization, development economics, and applied econometrics. Lei conducts an interdisciplinary, policy-oriented, and data-intensive research program at the nexus of energy, environment, public health, science, and innovation. His research themes include investigating firm strategies, consumer behavior, and government policies during energy transitions; exploring institutions and policies impacting innovation, such as patent systems and university research licensing; and applying data mining, learning, and modeling techniques to large datasets related to science, energy, environment, and public health. His work has been published in top journals across multiple fields, and he has received research support from prominent agencies including the NSF, NIH, DOE, and USPTO.

Research topics

• Computer Science
• Artificial Intelligence
• Machine Learning
• Computer vision
• Physics
• Operating system

Selected publications

• Bridging the Gap Between Anchor-Based and Anchor-Free Detection via Adaptive Training Sample Selection

  2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) · 2020 · 2206 citations

  • Computer Science
  • Computer Science
  • Artificial Intelligence

  Object detection has been dominated by anchor-based detectors for several years. Recently, anchor-free detectors have become popular due to the proposal of FPN and Focal Loss. In this paper, we first point out that the essential difference between anchor-based and anchor-free detection is actually how to define positive and negative training samples, which leads to the performance gap between them. If they adopt the same definition of positive and negative samples during training, there is no obvious difference in the final performance, no matter regressing from a box or a point. This shows that how to select positive and negative training samples is important for current object detectors. Then, we propose an Adaptive Training Sample Selection (ATSS) to automatically select positive and negative samples according to statistical characteristics of object. It significantly improves the performance of anchor-based and anchor-free detectors and bridges the gap between them. Finally, we discuss the necessity of tiling multiple anchors per location on the image to detect objects. Extensive experiments conducted on MS COCO support our aforementioned analysis and conclusions. With the newly introduced ATSS, we improve state-of-the-art detectors by a large margin to 50.7% AP without introducing any overhead. The code is available at https://github.com/sfzhang15/ATSS.

  DOI

• UA-DETRAC: A new benchmark and protocol for multi-object detection and tracking

  Computer Vision and Image Understanding · 2020 · 674 citations

  • Computer Science
  • Artificial Intelligence
  • Computer Science
  DOI

Frequent coauthors

• Stan Z. Li

  428 shared

• Xiangyu Zhu

  China State Shipbuilding (China)

  396 shared

• Jun Wan

  255 shared

• Shengcai Liao

  98 shared

• Zichang Tan

  95 shared

• Jianzhu Guo

  79 shared

• Yi Dong

  XinHua Hospital

  76 shared

• Yang Yang

  University of Electronic Science and Technology of China

  74 shared

Awards & honors

• NSF Award: CIVIC-FA Track A: Leveraging Existing Fiber-Optic…
• NSF Award: III:Small: Learning Latent Representations of Het…
• Repurposing Center for Energy Transition (ReCET)

Similar researchers at Pennsylvania State University

• Long-Qing Chenh-150
• Qin Wangh-123
• Richard Alleyh-120
• William H. Bruneh-99
• Ayusman Senh-96