Datasets:

claytonsds
/

UTM_Detection_Dataset

license: apache-2.0
task_categories:
  - image-segmentation
tags:
  - object-detection
  - u-tube-manometer
  - gauge-reader
  - water-column
language:
  - pt
  - en
pretty_name: U-Tube Manometer Detection Dataset 2025-2026
size_categories:
  - 1K<n<10K

Dataset Description

This dataset was created to support Article X and is available on Roboflow and Hugging Face.

It consists of images of U-tube manometers constructed using a transparent PVC water level hose (5/16" × 1 mm) and flexible measuring tapes of different colors, each with a length of 150 cm (60 inches). The manometric fluids represented in the dataset include water, oil, and dyed water, and the dataset is intended for computer vision object detection tasks, specifically targeting the detection of liquid column levels.

The dataset includes 8889 images, and all annotations were performed using the RoboFlow Annotate Tool.

Dataset Split

Training set: 6253 images (70%)
Validation set: 1758 images (20%)
Test set: 878 images (10%)

Data Structure

The dataset follows a standard split structure:

dataset/
├── train/
│   ├── images/
│   │   ├── image5214.jpg
│   │   └── ...
│   └── labels/
│       ├── image5214.txt
│       └── ...
├── valid/
│   ├── images/
│   │   ├── image3710.jpg
│   │   └── ...
│   └── labels/
│       ├── image3710.txt
│       └── ...
└── test/
│   ├── images/
│   │   ├── image18313.jpg
│   │   └── ...
│   └── labels/
│       ├── image18313.txt
│       └── ...
└── data.yaml

The dataset includes TXT annotations and a YAML configuration file compatible with YOLOv8.

License

This dataset is released under the Apache License 2.0.

Citation

@misc{utm_detection_dataset,
    author    = {Santos, Clayton Silva},
    title     = {{UTM} {Detection Dataset}},
    year      = {2026},
    month     = {feb},
    publisher = {Roboflow},
    version   = {12},
    doi       = {10.57967/hf/7783},
    url       = {http://universe.roboflow.com/visiongauge/utm_dataset_detection/dataset/12}
}