Parking-Spot-Occupancy-Recognition

Toward Parking Spot Occupancy Recognition: A Self-Supervised Approach

This repository contains the models and evaluations associated with the paper Toward Parking Spot Occupancy Recognition: A Self-Supervised Approach.

The official code is available on GitHub: Parking-Spot-Occupancy-Recognition.

Overview

This work proposes a self-supervised approach for parking spot occupancy recognition that requires no labeled samples from the target parking lot. The training pipeline consists of three stages:

1. Stage 1: Self-supervised pretraining using the SimCLR framework on generic data (ImageNet).
2. Stage 2: Self-supervised fine-tuning on parking spot occupancy data (PKLot, CNRPark-EXT, and PLds) to adapt to specific characteristics of parking spot images.
3. Stage 3: Supervised fine-tuning on parking spot occupancy data to refine the model's ability to classify parking spot occupancy accurately.

Evaluation

The models are evaluated under a leave-one-out cross-environment protocol on three public datasets: PKLot, CNRPark-EXT, and PLds. For more details on the performance and implementation of the Strong General Model and Specialized Model, please refer to the paper.

Citation

@article{kujavski2026toward,
  title={Toward Parking Spot Occupancy Recognition: A Self-Supervised Approach},
  author={Kujavski, Luan Marko and Laroca, Rayson and de Almeida, Paulo Lisboa},
  journal={arXiv preprint arXiv:2606.20886},
  year={2026}
}