defect detection

🔬  MicroStruct-VQA-1K 🔬

MicroStruct-VQA -1K is a synthetic vision-language benchmark designed to advance multimodal learning for metallurgical microstructure analysis.

Post-harvest fruit quality inspection in industrial conveyor environments is constrained by the absence of realistic video datasets. Existing benchmarks predominantly consist of static images collected under controlled conditions, lacking the temporal dynamics, full-surface coverage, and industrial-grade annotations needed for real-world deployment.

The GC10-DET dataset is a real industrial surface defect dataset, comprising 3570 high-resolution (2048 x 1000) grayscale images with ten types of surface defects. These are Punching hole (Pu), Welding line (Wl), Crescent gap (Cg), Water spot (Ws), Oil spot (Os), Silk spot (Ss), Inclusion (In), Rolled pit (Rp), Crease (Cr), and Waist folding (Wf). These defects are naturally occurring and exhibit a variety of morphologies. After preprocessing the original dataset, the dataset used in the experiment consisted of 2470 samples, of which we used 2223 for training and 247 for model validation.

This is part of the ultrasound image dataset used in the paper "BRPLE: Bayesian Regularized Post-hoc Local Explanations".

Casting Billet: This dataset comprises 1,060 images with resolutions ranging from 96×106 to 3,228×492, depicting high-temperature continuous casting billets. The images were acquired using an advanced optical inspection system comprising two 4096-pixel line-scan CCD cameras equipped with blue laser line illumination, achieving a spatial resolution of 0.18 mm/pixel. Among these, 780 images contain surface defects, categorized into six types: scratch (Sc), weld slag (WS), cutting opening (CO), water slag mark (WSM), slag skin (SS), and longitudinal crack (LC).

This dataset is related to the paper [3D CT Slice Image-Based Algorithm for Non-Wet Defect Inspection in Solder Joints]. 

Please refer to the following link: here

Due to the lack of publicly available injection-molded product defect datasets and the diversity of defects in terms of shapes, sizes, and textures, we collects defect samples from injection molding factories to ensure the model performs well in real industrial scenarios. To ensure the quality and usability of the data, after analyzing the sample data, data cleaning is performed to remove the irregular images.

<p>The dataset forwas collected by UAVs equipped with camera heads to capture images of insulators on power transmission lines. These images have a resolution of 3872×2592 pixels. A total of 488 insulator defect images were selected, and the data was annotated using the LabelMe annotation software. This study's dataset annotated four types of labels: insulator, damaged, Flashover, and hammer. The insulator is a positive class label, and damaged, Flashover, and hammer are negative class labels.

</p>

The steel tube dataset comprises comprehensive information on various attributes related to steel tubes, encompassing dimensions, material composition, manufacturing processes, and performance characteristics. This dataset facilitates in-depth analysis of steel tube properties, aiding researchers, engineers, and industry professionals in optimizing designs, ensuring structural integrity, and advancing materials science in the context of steel tube applications.