Lpipe Dataset

📄 DualMatch: A Dual EMA Teacher for Underwater Semi-Supervised Pipeline Segmentation

The Lpipe dataset is an underwater image dataset designed for semantic segmentation tasks involving subsea environments.
It includes manually annotated RGB images containing pipelines, humans, animals, and robots.
The dataset aims to support research in underwater computer vision, autonomous robotics, and oil and gas inspection systems.

---

Dataset Summary

The Lpipe dataset provides 716 underwater images paired with 716 pixel-level segmentation masks.
Each mask is color-coded in RGB, where each color corresponds to one of the four semantic classes:

Class ID	Class Name	Description
0	Background	Underwater background and seabed
1	Pipeline	Subsea pipeline or tubular structures
2	Human	Diver or human presence
3	Animal	Marine animals (fish, crustaceans, etc.)
4	Robot	Underwater inspection or maintenance robot

This dataset was created to assist in the development and evaluation of semantic segmentation models for complex underwater scenarios.

---

Supported Tasks and Benchmarks

The dataset can be used for:

• Semantic Segmentation
• Semi-supervised Segmentation (splits available on GitHub)
• Object detection and domain adaptation in underwater imagery

---

Dataset Structure

Inside the dataset, two folders are provided:

Lpipe/

• images/ # RGB underwater images (.jpg)
• masks/ # Segmentation masks (.png, RGB-coded)

Although the dataset on Hugging Face includes only images and masks,
official training/validation/test splits (for semi-supervised learning) are available in the associated GitHub repository:
🔗 https://github.com/EduardoLawson1/DualMatch

---

Data Fields

Field	Type	Description
image	RGB Image (.jpg)	The original underwater image
mask	RGB Image (.png)	Pixel-level color-coded segmentation mask

---

Data Splits

There are 716 total samples.
The dataset is not pre-split in the Hugging Face version, but the following splits are defined in the GitHub repository:

• train.txt
• val.txt
• test.txt
• unlabeled.txt (for semi-supervised training)

---

Licenses

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.
You are free to share, copy, redistribute, adapt, and build upon the material for any purpose, provided that you give proper credit to the original author.

---

Citation

If you use this dataset, please cite the related publication:

bibtex @article{silvadualmatch, title={DualMatch: A Dual EMA Teacher for Underwater Semi-Supervised Pipeline Segmentation}, author={Silva, Eduardo L and Schein, Tatiana T and Briao, Stephanie L and Anastacio, Gabriel L and Oliveira, Felipe G and Drews-Jr, Paulo LJ} }