Semi-Supervised Video Object Segmentation for Electric Vehicle Charger Socket Tracking

Objective

Extended previous Electric Vehicle Charger Socket Instance Segmentation and Electric Vehicle Charger Socket Instance Segmentation & Tracking projects by building a custom DAVIS-style dataset with YOLOv8-Seg and adapting it for XMem to enable temporally consistent pixel-level segmentation across video frames.

Stack

Delivery highlights

• Developed a semi-supervised video object segmentation system for electric vehicle charger sockets by extending previous Electric Vehicle Charger Socket Instance Segmentation and Electric Vehicle Charger Socket Instance Segmentation & Tracking projects into temporal video segmentation. Built a custom DAVIS-style dataset generation pipeline from electric vehicle charging videos using YOLOv8-Seg to generate frame-level binary masks, and adapted the dataset structure, annotation format, and frame-mask sequences to support XMem training and inference. Designed video preprocessing, frame extraction, mask generation, and sequence construction workflows to create temporally aligned data for video object segmentation. Leveraged XMem for memory-based mask propagation to preserve temporal consistency and stable object identity across frames, providing fine-grained pixel-level segmentation continuity that is more suitable for video object segmentation than box-level tracking methods such as ByteTrack.