Hardware Tools Complete Dataset

This repository provides a comprehensive, multi-device image dataset covering 24 fine-grained workshop hardware and hand tool categories. Images were captured directly in practical, uncontrolled workshop environments using three distinct smartphone camera modules (OnePlus Nord CE 4, Nothing CMF Phone 2 Pro, and Nothing Phone 2) to capture natural variations in viewpoint, background clutter, ambient/artificial lighting, and optical response. All target objects are annotated with precise 2D bounding boxes in standard COCO JSON format. The master archive (ML_Hardware_Tools_Complete_Dataset.zip) contains: 1. ML_Hardware_Tools_Raw_COCO/: The 4,545 original, unaugmented workshop images partitioned into train, validation, and test splits for custom preprocessing, domain adaptation, and few-shot or self-supervised learning. 2. ML_Hardware_Tools_Augmented_COCO/: The standardized 12,150-image benchmark dataset (10,140 train, 1,006 validation, 1,004 test) containing 13,306 annotated instances used to establish object detection baselines (YOLO variants and RF-DETR). 3. Dataset_Documentation_&_User_Guide.txt: Complete documentation detailing class taxonomy (24 classes), folder organization, annotation schemas, and citation instructions. Source code and benchmark training pipelines are publicly available at: https://github.com/mark-0polo/ML-CODES.git