Industrial CT Images and Annotations of Iron Ore Pellets for Internal Defect Segmentation

Overview This dataset presents a high-resolution collection of industrial Computed Tomography (CT) images and their corresponding color-coded pixel annotations for iron ore pellets. It is specifically curated to support research in material phase segmentation, porosity analysis, and the development of computer vision algorithms for metallurgical characterization. Data Description Total Samples: 20 pairs of high-resolution images. Format: All files are provided in .tif format. Resolution: $5052 \times 4840$ pixels per image. Source Images: Enhanced industrial CT scans. These images have undergone specialized preprocessing (including contrast enhancement and noise reduction) to optimize the visibility of internal pellet structures and grain boundaries. Annotated Images: These images provide human-verified, pixel-level labeling overlaid on the original scans. Unlike binary masks, these annotations retain the original image context while using specific colors to distinguish different material phases. Naming Convention To facilitate automated data loading and pairing, a strict naming convention is applied: Source Image: [Index_out].tif (e.g., t790_out.tif) Annotated Image: seg_[Index].tif (e.g., seg_t790_out.tif) Phase Identification & Color Scheme Three distinct phases within the pellets are identified using the following RGB color coding: Red (R:255, G:0, B:0): Pores (including micro-pores and internal cracks). Green (R:0, G:255, B:0): Liquid Phase (binding phase formed during induration). Blue (R:0, G:0, B:255): Hematite (primary iron-bearing mineral phase). Potential Applications Training and evaluation of semantic segmentation models (e.g., U-Net, ResNet-based architectures). Quantitative characterization of pellet microstructure and porosity. Benchmarking image enhancement and phase identification algorithms in mineral processing.