M2 screw defect detection dataset

The collection specifically focuses on isolating anomalies across two distinct regions: the screw Body (thread) and the screw Head. The dataset is divided into structured folders to support binary classification tasks (OK for compliant components and NG for defective components) across both training and testing pipelines.

The training data has been strictly class-balanced to mitigate algorithmic optimization bias, while the testing data provides an independent validation benchmark for evaluating real-world system generalization. Total Image Count: 5,536 images Dataset distribution as follows M2_Screw_Dataset/ ├── Body/ │ ├── Train/ │ │ ├── OK/ (1,000 images) │ │ └── NG/ (1,000 images) │ └── Test/ │ ├── OK/ (500 images) │ └── NG/ (500 images) └── Head/ ├── Train/ │ ├── OK/ (1,000 images) │ └── NG/ (1,000 images) └── Test/ ├── OK/ (257 images) └── NG/ (279 images)