GTCrack

Dataset Description

Automated detection of expressway pavement cracks is critically important for infrastructure inspection and maintenance. In this task, range information provides a key geometric cue for distinguishing genuine cracks from pavement textures, shadows, stains, and other visual interference. However, industrial-grade pavement data acquisition equipment is expensive, and pixel-wise annotation is labor intensive. Consequently, most existing pavement crack datasets contain only visual imagery and cannot precisely represent the geometric morphology of the pavement surface.

GTCrack was developed to alleviate this data bottleneck. A survey vehicle equipped with two laser line profilers was used to collect approximately 16 km of three-dimensional point-cloud data on expressways. The point clouds were projected into 4,808 paired two-dimensional samples. Each sample contains an intensity image describing surface appearance and a range image describing surface geometry, allowing the two heterogeneous modalities to jointly capture pavement texture and geometric morphology.

From the collected data, 683 crack-containing samples were selected for manual annotation. Two master's students independently produced pixel-wise crack masks, after which a Ph.D. candidate reviewed and refined the annotations. The GTCrack dataset contains 683 complete intensity-range-mask triplets and preserves the provided train, validation, and test partition.

Dataset Structure

The dataset is distributed in Hugging Face ImageFolder format. Every record contains two aligned input modalities and one semantic segmentation target.

Columns

• image: grayscale intensity image representing pavement surface appearance.
• range_image: grayscale range image representing pavement surface geometry.
• label: binary pixel-wise crack mask stored as an 8-bit grayscale PNG.
• id: sample identifier.
• width, height: image dimensions.

All three image fields are spatially aligned and have a resolution of 472 x 1000 pixels. Mask class IDs are 0 for background and 1 for crack.

Dataset Splits

Split	Samples
Train	478
Validation	102
Test	103
Total	683

Annotation Quality Control

The masks were independently annotated by two master's students. A Ph.D. candidate subsequently reviewed the annotations and refined ambiguous or inconsistent regions to improve pixel-level label quality.

Intended Use

GTCrack is intended for research on pavement crack semantic segmentation, intensity-range multimodal fusion, geometry-aware crack recognition, and the robust discrimination of cracks from texture and illumination interference.

Loading

from datasets import load_dataset

ds = load_dataset("hf-gx/GTCrack")
print(ds)
print(ds["train"].features)

sample = ds["train"][0]
image = sample["image"]
range_image = sample["range_image"]
mask = sample["label"]

---

GTCrack 中文说明

数据集简介

高速公路路面裂缝的自动检测对基础设施巡检与养护具有重要意义。在裂缝识别过程中， 距离信息能够提供关键的几何线索，有助于区分真实裂缝与路面纹理、阴影、污渍及其他 视觉干扰。然而，工业级路面数据采集设备成本较高，像素级标注工作又十分耗时，导致 同时包含外观和几何信息的公开路面裂缝数据集较为缺乏。现有研究因此大多只依赖视觉 图像，难以准确描述路面的几何形态。

为缓解这一数据瓶颈，我们使用配备两台激光线轮廓仪的道路检测车，在高速公路上采集 了约 16 km 的三维点云数据。点云数据被投影为 4,808 组二维配对样本，每组样本同时 包含强度图和距离图。其中，强度图反映路面表观与纹理信息，距离图反映路面几何形态， 两种异构模态可共同描述路面表面特征。

研究人员从采集数据中筛选出 683 个含裂缝样本。两名硕士研究生分别完成像素级裂缝 标注，随后由一名博士研究生对标注结果进行审核和修订。GTCrack 数据集包含 683 组 文件完整且严格配准的“强度图-距离图-分割掩膜”样本，并保留原始的训练集、验证集 和测试集划分。

数据结构

本数据集采用 Hugging Face ImageFolder 格式组织。每条记录包含两个空间对齐的输入 模态和一个语义分割标签。

字段说明

• image：灰度强度图，用于描述路面表观与纹理信息。
• range_image：灰度距离图，用于描述路面表面几何形态。
• label：像素级二值裂缝分割掩膜，以 8 位灰度 PNG 保存。
• id：样本唯一标识。
• width、height：图像宽度和高度。

三个图像字段均已严格空间对齐，分辨率统一为 472 x 1000 像素。掩膜类别编号为： 0 表示背景，1 表示裂缝。

数据集划分

数据划分	样本数
训练集	478
验证集	102
测试集	103
合计	683

标注质量控制

裂缝掩膜由两名硕士研究生独立标注，再由一名博士研究生对存在歧义或不一致的区域进行 审核和修订，以提高像素级标签的准确性与一致性。

适用范围

GTCrack 可用于路面裂缝语义分割、强度图与距离图多模态融合、几何感知裂缝识别，以及 复杂纹理和光照干扰条件下的鲁棒裂缝检测研究。

加载方式

from datasets import load_dataset

ds = load_dataset("hf-gx/GTCrack")
print(ds)
print(ds["train"].features)

sample = ds["train"][0]
image = sample["image"]
range_image = sample["range_image"]
mask = sample["label"]