TR-MobileNetV3: A Lightweight Neural Network for Tree-Ring Segmentation
Description
This deposit accompanies the manuscript TR-MobileNetV3: A Lightweight Neural Network for Tree-Ring Segmentation and provides materials related to the mobile deployment of the proposed model. Contents TR-MobileNetV3.rar — Project source code for the proposed TR-MobileNetV3 model, including training/inference scripts, configuration files, and the released model weights. Android application (“HLF Ring Detector”) — A client-side app for offline tree-ring image segmentation based on TR-MobileNetV3. It uses on-device inference (ONNX Runtime for Android), supports capture or gallery input, an adjustable binarisation threshold (0.0–1.0, default 0.5), and export of a segmentation overlay and a binary mask for follow-up analysis (e.g. ring-width or latewood proportion). Minimum platform: Android 7.0 (API 24). Sample images — A compressed archive of representative photographs illustrating typical inputs and use cases discussed in the paper (e.g. standard stem or disc cross-sections, field or museum settings, and more challenging surfaces). These images are for documentation and demonstration only and are not a substitute for the full research image dataset. Scope and limitations The segmentation model was trained primarily on Larix mastersiana latewood masks. Performance on other species or imaging conditions may vary; the images in Sample images are illustrative. For the complete experimental dataset, splits, and quantitative results, see the article and contact the authors if further research data are needed. Suggested citation If you use these materials, please cite the associated publication once available.