Lionel Ang

Key Skills

Software

CVAT

Top Subject Matter

No subject matter listed

Top Data Types

Image

Text

Video

Top Task Types

Action Recognition

Emotion Recognition

Text Generation

Translation Localization

Freelancer Overview

Creating high-quality datasets that enhance the precision and safety of autonomous vehicles in controlled environments. My expertise includes annotating sensor data (LiDAR, radar, and cameras) for object detection, path planning, and obstacle avoidance, with a focus on edge cases unique to low-speed applications like warehouse robots, last-mile delivery vehicles, and industrial automation. I have hands-on experience in domain-specific labeling challenges, such as fine-grained classification of pedestrians, cyclists, and static obstacles in complex, low-speed scenarios. I’ve contributed to projects involving semantic segmentation for indoor autonomous tugs and bounding-box annotation for curbside delivery bots, ensuring robust model performance in real-world settings. My rigorous attention to detail, familiarity with labeling tools (e.g., CVAT, Labelbox), and understanding of SAE autonomy levels (L2-L4) enable me to deliver training data.

Labeling Experience

Image recognition for low speed autonomous delivery

CVATImageBounding BoxPolygon

Led the annotation of multimodal sensor data (primarily camera images) for a low-speed autonomous vehicle system, focusing on urban last-mile delivery robots. The project involved labeling 20,000+ high-resolution images captured in diverse lighting/weather conditions to train perception models for object detection, lane marking recognition, and pedestrian behavior prediction. Tasks: Bounding Boxes: Annotated vehicles, pedestrians, cyclists, and static obstacles (e.g., traffic cones, dumpsters) with tight alignment for precise localization. Semantic Segmentation: Labeled drivable surfaces, sidewalks, and crosswalks pixel-wise to support path-planning algorithms. Polygons: Detailed annotations for irregular objects (e.g., construction barriers, partially visible assets) to minimize occlusion errors. Inter-annotator Agreement (IAA): Maintained >95% consistency across a 5-annotator team via CVAT’s review workflows and overlap assignments.

Led the annotation of multimodal sensor data (primarily camera images) for a low-speed autonomous vehicle system, focusing on urban last-mile delivery robots. The project involved labeling 20,000+ high-resolution images captured in diverse lighting/weather conditions to train perception models for object detection, lane marking recognition, and pedestrian behavior prediction. Tasks: Bounding Boxes: Annotated vehicles, pedestrians, cyclists, and static obstacles (e.g., traffic cones, dumpsters) with tight alignment for precise localization. Semantic Segmentation: Labeled drivable surfaces, sidewalks, and crosswalks pixel-wise to support path-planning algorithms. Polygons: Detailed annotations for irregular objects (e.g., construction barriers, partially visible assets) to minimize occlusion errors. Inter-annotator Agreement (IAA): Maintained >95% consistency across a 5-annotator team via CVAT’s review workflows and overlap assignments.

2018 - 2019

Education

Work History