WoodChip-Detection: A Public Dataset for Dense Wood Chip Detection and Instance Segmentation

WoodChip-Detection is a publicly available image dataset for dense wood chip detection and instance-level segmentation in industrial biomass processing environments. The dataset is designed to support benchmarking of object detection, instance segmentation, and geometry-aware analysis methods under realistic, cluttered conditions. The dataset contains 223 RGB images with a total of 6,879 manually annotated wood chip instances, collected across three independent acquisition sources, one of which is further divided into two batches representing distinct capture sessions. Images were acquired using a fixed laboratory imaging setup with consistent illumination and camera viewpoint, while preserving natural variability in chip arrangement, density, and appearance. All wood chip instances are annotated manually using fine-grained polygon masks in the LabelMe format. A single semantic class is used, corresponding to individual wood chips. The polygon annotations support instance-level segmentation and geometry analysis, and can be converted to bounding boxes for compatibility with standard object detection frameworks such as DETR, Faster R-CNN, RetinaNet, and YOLO, as well as segmentation-based models. Image resolutions are not uniform and vary across sources and batches, reflecting practical industrial imaging conditions rather than artificially standardized settings. The dataset exhibits a wide range of instance densities, including both sparse and highly cluttered scenes, making it suitable for evaluating detector and segmenter robustness, small-object localization, and density-dependent performance. Metadata is provided per acquisition source (and per batch where applicable), including dataset-level summary statistics and per-image instance counts. This dataset supports research in: object detection instance-level segmentation geometry and size analysis robustness under clutter and domain shift industrial visual inspection and biomass material characterization Detailed model benchmarking and performance results are given in an accompanying journal paper: Eskorouchi, A., Rahman, A., Street, J. T., Marufuzzaman, M., & Wang, H. (2025, June). Enhanced DETR-Based Framework for Automated Wood Chip Size Distribution Estimation in High Volume Biomass Manufacturing. In International Conference on Flexible Automation and Intelligent Manufacturing (pp. 540-548). Cham: Springer Nature Switzerland.Other studies related to the dataset are as follows: Eskorouchi, A., Rahman, A., Marufuzzaman, M., Street, J. T., & Wang, H. Automated Wood Chip Size Estimation with YOLO-Based Object Detection. If you use the dataset, please cite the dataset or the journal article above. Thank you. Acknowledgements: This work is supported by the Sustainable Bioeconomy through Biobased Products and Engineering for Agricultural Production and Processing programs, project award no. 2020-67019-30772 and 2022-67022-37861, from the U.S. Department of Agriculture’s National Institute of Food and Agriculture. Any opinions, findings, conclusions, or recommendations expressed in this publication are those of the author(s) and should not be construed to represent any official USDA or U.S. Government determination or policy.