Research

This video demonstrates how light detection works in real-world scenarios.

View More Details
Light Detection
Accurate detection of these light sources is crucial for recognizing objects, predicting behaviors, and understanding the environment, particularly in the context of autonomous driving and surveillance systems.
  • Published Paper
  • A New Multi-Source Light Detection Benchmark and Semi-Supervised Focal Light Detection
  • Conference : NeurIPS 2024
  • Authors : Jae-Yong Baek, Yong-Sang Yoo and Seung-Hwan Bae*
  • Multi-Source Light Detection with Semi-supervised Learning and Lightness Focal Loss
  • Conference : 제 36회 영상처리 및 이해에 관한 워크샵
  • Authors : Yong-Sang Yoo, Jae-Yong Baek and Seung-Hwan Bae*
  • Uses Cases
  • ● Autonomous driving
  • ● Night vision
  • ● Industrial Automation and Robots

This video demonstrates how On Board AI works in real-world scenarios.

View More Details
On Board AI
On-Board AI refers to performing AI computations directly on the device, enabling real-time processing. Jetson, developed by NVIDIA, is a GPU-based platform ideal for high-performance tasks like robotics and autonomous driving. Hailo-8 is a low-power AI chip designed for fast inference in lightweight devices like CCTV and IoT sensors. Jetson is developer-friendly and versatile, while Hailo focuses on energy efficiency and speed. Both platforms support AI without relying on the cloud, making them key technologies for Edge AI.
  • Published Paper
  • 군용 도메인 영상에 대한 서버와 온-보드 간의 객체 검출 성능 분석
  • Journal : 한국컴퓨터정보학회논문지
  • Authors : Du-Hwan Hur, Dae-Hyeon Park, Deok-Woong Kim, Jae-Yong Baek, Jun-Hyeong Bak and Seung-Hwan Bae*
  • Uses Cases
  • ● Drones & UAV Systems & Robotics
  • ● Healthcare & Wearables
  • ● Smartphones & IoT Devices

This video demonstrates how Data-Free Quantization works.

Data-Free Quantization
Data-Free Quantization is a model compression technique that reduces the precision of neural network weights and activations without requiring access to the original training data, typically by generating synthetic data or using statistical information to preserve model performance.
  • Published Paper
  • 쌍방향 지식 전이를 통한 데이터 프리 양자화
  • Conference : 제 37회 영상처리 및 이해에 관한 워크샵
  • Authors : Deok-Woong Kim, Chan-Seop Park, Jae-Yong Baek, and Seung-Hwan Bae*
  • 생성 모델을 이용한 데이터 프리 양자화를 위한 Bit-width Aware Generator와 채널 어텐션 기반 중간 레이어 지식 증류
  • Journal : 한국컴퓨터정보학회논문지
  • Authors : Jae-Yong Baek, Dae-Hyeon Park, Yong-Sang Yoo, Du-Hwan Hur, Deok-Woong Kim, Hyuk-Jin Shin and Seung-Hwan Bae*
  • Uses Cases
  • ● Drones & UAV Systems & Robotics
  • ● Healthcare & Wearables
  • ● Smartphones & IoT Devices

This video demonstrates how Multi-Modal Tracking works.

Multi-Modal Tracking
Multi-Modal Tracking is the process of simultaneously using and integrating information from multiple sensor modalities—such as RGB images, thermal cameras, depth sensors, or LiDAR—to robustly track objects over time, even in challenging environments.
  • Published Paper
  • ASTR: Efficient Multi-Modal Tracking with Asymmetric Transformer
  • Conference : 제 37회 영상처리 및 이해에 관한 워크샵
  • Authors : Jeong-Hun Ha, Dae-Hyeon Park, and Seung-Hwan Bae*
  • Uses Cases
  • ● Autonomous driving
  • ● Night vision
  • ● Industrial Automation and Robots

This video demonstrates how 3D Object Detection works in real-world scenarios.

View More Details
3D Object Detection
3D Object Detection is a computer vision task that involves identifying, localizing, and estimating the size and orientation of objects in three-dimensional space using input data such as images, LiDAR point clouds, or RGB-D information.
  • Published Paper
  • Region-Aware Knowledge Distillation between Monocular Camera-based 3D Object Detectors
  • Journal : ICT Express
  • Authors : Se-Gwon Cheon, Hyuk-Jin Shin and Seung-Hwan Bae*
  • 3D Object Detection via Multi-Scale Feature Knowledge Distillation
  • Journal : 한국컴퓨터정보학회논문지
  • Authors : Se-Gwon Cheon, Hyuk-Jin Shin and Seung-Hwan Bae*
  • 단안 카메라 기반의 3D 객체 검출을 위한 샘플 인식 지식 증류
  • Conference : 제 37회 영상처리 및 이해에 관한 워크샵
  • Authors : Mina Baek, Se-Gwon Cheon, Hyuk-Jin Shin and Seung-Hwan Bae*
  • Uses Cases
  • ● Autonomous driving
  • ● Drones & UAV Systems & Robotics

This video demonstrates how Multi-Object Tracking works.

View More Details
Multi-Object Tracking
The goal of multi-object tracking is to estimate the states of multiple objects, such as locations, velocities, and sizes, while conserving their identifications. We developed confidence-based tracking and online appearance learning, and deep learning algorithms to solve this problem.
  • Published Paper
  • Decode-MOT: How Can We Hurdle Frames to Go Beyond Tracking-by-Detection?
  • Journal : IEEE Transactions on Image Processing (TIP)
  • Authors : Seong-Ho Lee, Dae-Hyeon Park, and Seung-Hwan Bae*
  • Effective Multi-Object Tracking via Global Object Models and Object Constraint Learning
  • Journal : Sensors
  • Authors : Yong-Sang Yoo, Seong-Ho Lee and Seung-Hwan Bae*
  • RT-MOT: Confidence-Aware Real-Time Scheduling Framework for Multi-Object Tracking Tasks
  • Conference : IEEE Real-Time Systems Symposium
  • Authors : Donghwa Kang, Seunghoon Lee, Hoon Sung Chwa, Seung-Hwan Bae, Chang Mook Kang, Jinkyu Lee and Hyeongboo Baek
  • Uses Cases
  • ● Autonomous driving
  • ● Night vision
  • ● Industrial Automation and Robots
VL Lab. © 2021 All Right Reserved