EV Charger Detection
Industrial AI / Computer Vision
An end-to-end computer vision pipeline for collecting, filtering, labeling, and classifying EV charger imagery.
Overview
The project combines web-collected EV charger imagery with automated labeling, embedding-based filtering, and model training.
The problem
Image collections can contain uncertain labels, near duplicates, and source-related leakage that undermine model evaluation.
My contribution
Built the pipeline around YOLO auto-labeling, ResNet50 embeddings, human review, and leakage-aware dataset splitting.
System or workflow
- 01Collect candidate imagery.
- 02Generate initial labels with YOLO.
- 03Filter candidates using pretrained embeddings and cosine similarity.
- 04Review uncertain samples and train the classifier on verified data.
Technical highlights
- 01Human-in-the-loop annotation refinement
- 02Cosine similarity filtering with pretrained embeddings
- 03Leakage-aware train and validation splitting by source URL