Trendspek Optimizes Structural Crack Detection with Datature

Datature is proud to support Trendspek’s mission of transforming asset management through its Precision Asset Intelligence 3D visualisation software.

Wei Loon Cheng

About Trendspek

Trendspek offers a purpose-built Precision Asset Intelligence-powered platform that is designed to process tens of thousands of high-quality images into exact 3D replicas known as Precision Reality Twins. This helps enterprises to accelerate and cut down costs for asset markup, monitoring, and maintenance planning through this virtual medium.


Smart City, Construction, Property, Water, Maritime, Energy, Mining


Sydney, New South Wales, Australia

Trendspek’s Industry-Leading Asset Software 

Datature, a leading MLOps platform for computer vision technology, and Trendspek, an industry-leading 3D asset management and reporting software, have developed a proof-of-concept advanced machine learning model for structural crack inspection in what marks an exciting innovation for 3D asset management.

These advanced machine learning models are trained to identify and precisely segment cracks on structural facades, such as concrete silos. The prediction results from the Datature models are integrated with Trendspek Precision Reality Twins and rendered on Trendspek’s Precision Asset Intelligence platform to provide fast, accurate insights to users.

While manual analysis remains crucial for its domain expertise, Trendspek recognizes the potential for technology to enhance and complement human efforts in inspection processes. In collaboration with Datature, Trendspek aims to augment its existing platform capabilities with AI-powered analysis, offering users a tool to speed up crack defect identification by up to 5 times.

While still in its development, this integration aims to allow users to leverage the efficiency of Datature’s automated computer vision solutions while still retaining the valuable input and oversight of human expertise.

Datature’s Solutions

Datature is an end-to-end MLOps platform designed to facilitate users in creating and operationalizing a custom computer vision model pipeline with their own data without the need for code. Trendspek has leveraged our Nexus platform starting from the data collection stage, and swiftly developing working models in less than two weeks.

Project Dashboard on Datature Nexus showing statistics on assets and annotations.

Once drone image data had been imported from Trendspek’s platform, Trendspek was able to leverage Nexus’ in-house Annotator for the data annotation phase. Datature’s suite of manual annotation tools such as Paintbrush, together with smart labelling tools such as Intellibrush allowed Trendspek to precisely and efficiently label polygon segmentations around structural cracks.

Annotation of cracks using Datature Nexus' in-house annotation suite.

Datature’s Automation tool also empowered Trendspek to oversee annotation workflows for a team of annotators. The labelling team was able to collaboratively annotate and review annotations using separate user accounts, thereby greatly accelerating the annotation process and enhancing quality assurance pipelines.

Datature Nexus' Automation workflow for collaborative annotation and task delegation.

After the labelling process, the team at Trendspek was able to easily and rapidly train a robust semantic segmentation model within 24 hours. To quickly identify the small cracks present in the large, high-quality 8K images, Trendspek utilized Datature Nexus’ Sliding Window feature that was designed specifically to aid small object detection. Image augmentation techniques were also incorporated to help models become more robust against varying weather conditions. Within the first experimentation iteration, they were able to produce a working proof-of-concept model that can detect cracks in new, unseen images.

Datature Nexus' comprehensive training and validation metrics.
Datature Nexus' Advanced Evaluation for intuitive visualization of what the model is learning during the training process.

At the deployment level, Trendspek’s platform leveraged Datature’s Inference API to minimize the development necessary for a production-level deployment. Datature’s Inference API provided a no-code deployment setup on Nexus, provisioning fully-maintained, scalable, and dedicated GPUs on the cloud to host Trendspek’s segmentation model. Inference was easily achieved through simple API requests that can be directly invoked from Trendspek’s platform with the relevant image data. This allowed the Trendspek team to focus their efforts more on actionable insights such as on-platform rendering and any user alerts.

Tracking usage statistics of hosted model deployments on Datature's cloud servers.

Datature’s team of experienced domain experts, engineers, and developers also worked closely with Trendspek throughout the entire MLOps lifecycle in providing critical advice and support to help Trendspek achieve their goal of integrating a working model. Datature helped to develop specialized scripts aimed at addressing the technical difficulties associated with automating the crack detection process. The scripts included cross-platform compatibility support for pre-processing and post-processing data.

Rendering of cracks on the Precision Reality Twins on Trendspek's Precision Asset Intelligence platform. Cracks are automatically identified by models trained on Datature Nexus.

To facilitate future iterative development based on new data and changing conditions, Trendspek is also able to leverage Datature’s Active Learning capabilities and Management API to identify images with low-confidence predictions for re-uploading, model-assisted labelling of these images, and automated training and re-deployment of new models for production. Datature Nexus also facilitates the potential expansion of project scope for Trendspek, such as training new models to identify other structural defects.

Overall, Datature’s tools have allowed Trendspek to not only develop an effective computer vision solution to automatically detect structural cracks and other useful visual information for actionable insights, but also sustain a sophisticated and efficient pipeline to continuously improve on the model and overall user experience.

“Datature made it incredibly easy to annotate and train a model for reliable concrete crack detection in high resolution imagery. Datature's technical support was second to none, and we look forward to furthering our collaboration over coming months.”
- Josh Sinclair | Head of Product, Trendspek

Datature is proud to support Trendspek’s mission to automate and transform asset management through AI-powered inspection, and we are excited to see what’s next for Trendspek.

If you are an innovator looking to unlock the potential of deep learning to transform the Smart City industry, get in touch with us!

About Trendspek

Trendspek is the world’s leading Precision Asset Intelligence software that empowers you to virtually inspect and monitor built assets and infrastructure safely, securely and remotely. Generating high-fidelity Precision Reality Twin 3D models with up to 1mm detail, users can streamline defect identification and conduct thorough condition assessments from their desks. Launched 2018 as a solution to ageing assets and resource shortages, Trendspek’s secure cloud-based platform empowers you to collaborate, plan and manage assets from anywhere in the world to minimise risk and maximise certainty.

About Datature

Datature is an end-to-end MLOps platform for Computer Vision that allows teams and enterprises to build computer vision models without a single line of code. Teams can manage datasets, annotate, generate synthetic data, train and deploy - all in a single, secure cloud-based platform. With the rise of citizen data scientists, deep tech companies, and enterprises looking to adopt deep-learning - Datature equips these startups / experts with the tools required to build their own capabilities easily within weeks.

Build models with the best tools.

develop ml models in minutes with datature