Image Dataset for Intravenous (IV) Drop Detection with Dual Annotations (Bounding Box and Heatmap)

This dataset contains 3,458 high-resolution, annotated images designed to facilitate research and development in automated, computer vision-based monitoring of intravenous (IV) therapy. It was created to address the lack of diverse, publicly available datasets for the specific task of IV fluid drop detection. The images were captured under a wide range of realistic conditions, including varied ambient and artificial lighting, diverse clinical and non-clinical backgrounds, and multiple fluid types (e.g., clear saline, blood, and other colored solutions). The dataset also includes a significant number of negative samples (frames without drops) to enable the training of robust models that can avoid false positives from common artifacts like air bubbles or light reflections.The primary contribution of this dataset is its dual-annotation format, making it uniquely suitable for training and directly comparing two distinct detection paradigms:Bounding Box Labels: A YOLO-format .txt file is provided for each image, containing standardized bounding box coordinates for traditional object detection models.Heatmap Targets: A corresponding 26x26 NumPy (.npy) array is provided for each image, representing a ground-truth heatmap with a 2D Gaussian peak at the drop's center. This is ideal for training point-event or regression-based detection models.This dual format enables researchers to benchmark different algorithmic approaches on an identical, challenging, and diverse set of images. The data was used for the comparative study presented in the paper \A Dual-Module Computer Vision Approach for Real-Time IV Drip Rate and Fluid Level Monitoring.\

本数据集包含3458张高分辨率标注图像，旨在推动基于计算机视觉的自动化静脉输液（intravenous, IV）监测领域的研究与开发工作。本数据集的构建旨在解决当前静脉液滴检测特定任务领域缺乏多样化公开数据集的痛点。

所有图像均采集自多种真实场景，涵盖不同环境光与人工光照条件、多样的临床与非临床背景，以及多种输液液体类型（如澄清生理盐水、血液及其他有色溶液）。

本数据集还包含大量负样本（无液滴的图像帧），可用于训练能够规避气泡、光线反射等常见伪影引发假阳性的鲁棒模型。

本数据集的核心贡献在于其双标注格式，使其可独特适配两种不同检测范式的训练与直接对比：
1. 边界框标注：为每张图像提供YOLO格式的.txt文件，其中包含适用于传统目标检测模型的标准化边界框坐标。
2. 热图目标：为每张图像提供对应的26×26 NumPy（.npy）数组，该数组代表在液滴中心带有二维高斯峰值的真实基准热图，非常适合训练点事件检测或基于回归的检测模型。

这种双格式设计使研究人员能够在同一组兼具挑战性与多样性的图像集上，对不同算法方案开展基准测试。本数据集已用于论文《双模块计算机视觉方法实现实时静脉滴速与液位监测》（Dual-Module Computer Vision Approach for Real-Time IV Drip Rate and Fluid Level Monitoring）中的对比研究。