1. ABOUT THE DATASET -------------------- Title: Creator(s): Louis Masters [1] Organisation(s): 1. University of Leeds, School of Mechanical Engineering Rights-holder(s):Unless otherwise stated, Copyright 2025 University of Leeds Publication Year: 2025 Description: This dataset contains 5197 images, and the corresponding labels, of ceramic 3D printed parts. The parts were created using the bespoke CHAMP machine at The University of Leeds. Images were taken top-down after drying each layer. The images were used to train a Yolov8 model for defect detection purposes. The images labels correspond to two defect classes, under- and over-extrusions. The dataset is divided into training, validation, and test data. Cite as: Masters, Louis (2025) Dataset for 'Strategic Layer Reworking using Hybrid Additive Manufacturing for Defect-Free Ceramic Parts'. University of Leeds. [Dataset] https://doi.org/10.5518/1666. Related publication: Louis Masters, Dan Davie, Pablo J. Cevallos, Matthew P. Shuttleworth, Daniel Bara, James Warren, Mehmet Dogar, Robert Kay, Strategic Layer Reworking using Hybrid Additive Manufacturing for Defect-Free Ceramic Parts, Additive Manufacturing,2025,ISSN 2214-8604, https://doi.org/10.1016/j.addma.2025.104752. Contact: 2. TERMS OF USE --------------- Unless otherwise stated, this dataset is licensed under a Creative Commons Attribution 4.0 International Licence: https://creativecommons.org/licenses/by/4.0/. 3. PROJECT AND FUNDING INFORMATION ---------------------------------- Title: Defect Free Hybrid Additive Manufacturing Dates: 01/10/2021-30/9/2025 Funding organisation: EPSRC Grant no.: EP/T517860/1 4. CONTENTS ----------- File listing: The dataset contains three folders, train, val, test. Each folder contains sub-folders, of images (.jpg files), and labels (.txt files). A data.yaml file is included which details the file structure, and the defect classes required for training. 5. METHODS ---------- Images were captured using a high-resolution camera during the printing process. The images were labelled using Roboflow, and data augmentations were applied to the training images. Full details on the method to create this dataset are provided in the corresponding publication listed in section 1.