Jinlong Li Email & Phone Number

Palo Alto, California, United States

Personalization meets controllability in text-to-image diffusion.

Cleveland, Ohio, United States

A dedicated and results-driven researcher specializing in Computer Vision, Deep Learning, and Autonomous Driving. My expertise spans across pioneering deep learning technologies for vision-centric perception, with a keen focus on challenging areas such as 2D/3D object detection, domain adaptation, and the transformative use of Transformer and diffusion models.

Background: With the rising prevalence of vision-centric perception systems relying on camera sensors, addressing safety concerns associated with low-light conditions has become imperative for ensuring overall vehicle safety.Multi-Condition Diffusion Framework for Unpaired Low-Light Enhancement (CVPR2024 Rebuttal): Proposed a Diffusionmodel to enhance low-light camera images for autonomous driving, mitigating the need for extensive nighttime data collection and preserving daytime performance. Our method incorporates a novel multi-condition adapter that adaptively controls the input weights from different modalitiesto effectively illuminate dark scenes while maintaining context consistency.

Enhancing Cooperative Autonomous Vehicle (CAV) perception through V2V communication is crucial for improved detection performance. This project focused on real-world deployment, building dataset construction, Lossy Communication challenges, and addressing the domain gap between simulated and real data.Pioneering Dataset Construction (ICRA 2022 & CVPR 2023): Contributed to OPV2V, the first large-scale cooperative 3D LiDAR dataset, and as key contributor to V2V4Real, the first large-scale real-world V2V perception dataset.Cutting-edge Cooperative Perception Research under Lossy Communication (TIV 2023): Proposed the first re-search on V2V cooperative perception (point cloud-based 3D object detection) under lossy communication. Explored the impact of lossy communication on cooperative perception.Simulation-to-Reality Transfer Learning (ICRA2024): Proposed the first Simulation-to-Reality transfer learning framework for multi-agent cooperative perception using a novel Vision Transformer, named as S2R-ViT. Addressed Deployment Gap and Feature Gap between simulated and real data.Cross-Domain Learning for Multi-Agent Perception (ICRA2024): Proposed a novel Feature Distribution-aware Aggregation framework, which is the first research on multi-agent perception to address the Distribution Gap of different independent private data for training distinct agents.Adversarial GPS for Multi-Agent Attack (ICRA2024): Proposed the first research of adversarial GPS signals which are also stealthy for the V2V cooperative perception attacks, denoted as AdvGPS. Three statistically sensitive natural discrepancies in AdvGPS proposed to enhance the multi-agent perception attack in the black-box scenarios.

Addressing the challenge of limited labeled ground truths in nighttime images for deep learning modes, this project aimed to enhance vehicle perception in challenging driving scenarios, such as nighttime, and foggy weather. Leveraging transfer learning technology, the objective was tomaximize the use of labeled images to improve model performance.Situation-Sensitive Vehicle Detection Framework (TR-C 2021): Developed a framework for vehicle detection in both daytime and nighttime using labeled daytime images. Utilized CycleGAN as a style transfer technology to enhance model performance during nighttime conditions.Night-to-Day Translation for Vehicle Detection (TCSVT 2021): Introduced a detail-preserving Night-to-Day translation method for direct adaptation of daytime models to nighttime vehicle detection.Unsupervised Domain Adaptation for Adverse Conditions (WACV 2023): Proposed an unsupervised domain adaptation method for robust object detection in foggy and rainy conditions. Integrated AdvGRL and domain-level metric regularization for improved adaptability.

Objective: To develop methods enabling individuals with high tetraplegia to control aspects of helper robot reaching movements, incorporating motion planning, computer vision, and related technologies.Vision-Based Detection System Development: Led the development of a vision-based detection system for various foods within the Robotic-Assisted Feeding project. Defined food item characteristics to determine required actions, addressing different angles of food presentation. Implemented data augmentation for enhancedmodel generalization.

Doctor'S Degree, Computer Science

Experienced in pioneering deep learning technologies for computer vision and autonomous driving, with a proven track record.

Master'S Degree, Artificial Intelligence, Autonomous Driving, Intelligent Transportation System

Bachelor Of Transportation Engineering, Pavement Crack Detection