Xunpeng C.

Key Skills

Software

CVAT

Other

Top Subject Matter

Autonomous driving road-scene object detection

Social media sentiment classification (user comments)

Top Data Types

Image

Text

Top Task Types

Bounding Box

Classification

Freelancer Overview

Object Detection Annotation for Autonomous Driving Scenes. Core strengths include CVAT, Excel, and Other. Education includes Bachelor of Science, Wuxi Institute of Technology (2012). AI-training focus includes data types such as Image and Text and labeling workflows including Bounding Box and Classification.

Labeling Experience

Sentiment Classification Annotation for User Comments

TextClassification

Annotated user-generated social media comments to classify sentiment for downstream sentiment analysis model training. Labeled each comment as Positive, Negative, or Neutral and additionally tagged domain attributes such as product function feedback, logistics complaints, price complaints, and recommendation intention. Read comments in full, handled challenging cases like sarcasm/contrast using a guideline case library, and participated in second-round quality review to improve consistency. • Annotated 1,200+ Chinese user comments (~150/day on average). • Logged special cases (e.g., sarcastic statements) and reported them to refine the annotation manual. • Cross-checked consistency during a second-round quality review with other annotators. • Reached 97% quality-check consistency and delivered data directly to the model training pool.

2012

Object Detection Annotation for Autonomous Driving Scenes

CVATImageBounding Box

Performed large-scale road-scene object detection annotations for an autonomous driving R&D project, labeling vehicles, pedestrians, cyclists, traffic signs, and other entities according to fine-grained rules. Applied strict boxing requirements including tight object-edge fitting, no overlapping between boxes, and special handling for occluded pedestrians (over 50% occlusion marked separately). Filled attributes for each object such as moving/static and visibility (good/fair/poor), and conducted self-reviews to confirm uncertain annotations. • Annotated 400+ road-scene images with an average of 15–25 target objects per image. • Used CVAT rectangle boxing, label selection, attribute filling, and segment/annotation functions. • Ensured quality by marking uncertain boxes and consulting guidelines for confirmation. • Achieved automated quality-check pass with no rejections and delivered data for model training.

2012

Education

Work History