Artificial Intelligence Services
StrutFit
Aug 2019 – Oct 2023
At AIWORX Company, we take pride in offering specialized Artificial Intelligence services to StrutFit, a leading SaaS provider based in Auckland, New Zealand. StrutFit delivers innovative footwear size recommendations to renowned e-commerce businesses like Hoka, VivoBareFoot, and Bobux. Unlike its competitors who rely on indirect and time-consuming foot measurement techniques, StrutFit employs advanced AI and mobile photogrammetry for accurate and repeatable shoe size recommendations. Our goal was to enhance overall foot length accuracy and repeatability, while also incorporating foot width measurement as required by footwear manufacturers.
Our contributions to this project have been multifaceted and impactful. We spearheaded research and development initiatives, creating proof of concepts for foot width measurement, including metatarsal and aligned bounding width. We devised a novel foot measurement technique using plastic cards and transitioned from traditional machine learning techniques to powerful deep learning algorithms. Our team successfully incorporated foot rotation and image rejection mechanisms for high alignment angles, and we designed and deployed training and inference pipelines on AWS and GCP, complete with MLOps integration. This included memory-efficient object detection and subpixel semantic segmentation under variable lighting conditions, optimizing and fine-tuning models for higher accuracy, GPU memory usage, and loading time.
Additionally, we developed custom ELT pipelines for data cleaning and segregation in S3 buckets and adopted a microservices architecture. Our work also involved smartphone camera calibration and utilizing pre-estimated camera matrices and distortion coefficients. We enhanced accuracy using transfer learning and data augmentation with custom entity relationships in the images. Our comprehensive approach included code review and adherence to standard Agile practices, as well as HR training on deep learning, image segmentation, object detection, photogrammetry, AWS/GCP, and image labeling tools such as LabelBox and SuperAnnotate. Technologies employed in this journey included Python, C/C++, C#, TensorFlow, FastAPI, Flask, MLFlow, Deeplab, Detectron, Yolo, Unet, OpenCV, Keras, Pandas, Azure CICD, AWS ECR, ECS, SageMaker, EC2, S3, Lambda, CloudWatch, Docker, Terraform, Git, Slack, Asana, Notion, and Jira. Our collaboration with StrutFit exemplifies our commitment to driving innovation and excellence in AI solutions.