Vitoo AI Training Platform
Photometric stereo 2.5D is a surface inspection technique that captures multiple images of a stationary object under different lighting directions (typically 3 or more). By analyzing the brightness variations across these images, it computes the surface gradient (local slope) at each pixel, rather than absolute height. The result is a set of 2.5D feature maps — such as reflectance (albedo) map, gradient maps (surface tilt in X/Y directions), and curvature maps — which encode relative surface shape without full 3D coordinates. The "2.5D" term means the data represents height as a function of 2D image coordinates (like a 2D image with depth-like information), but it cannot represent overhanging or occluded structures. Traditional 2D imaging captures only grayscale or color intensity information under fixed, often uniform lighting. Its ability to detect defects relies entirely on color or contrast differences, which makes it poor for low contrast, textureless, or shiny surfaces. In contrast, photometric stereo 2.5D uses multiple directional lights that are turned on sequentially. It computes not just how much light is reflected, but the surface orientation (gradient) and curvature at each point. The output is not a single intensity image but several feature maps, including gradient maps, curvature maps, and a reflectance (albedo) map. As a result, 2.5D is excellent for detecting subtle raised or depressed defects (such as scratches, dents, or bumps) on uniform surfaces, whereas 2D imaging would miss them because they lack contrast. | Vitoo AI Training Platform is a graphical deep learning model training platform that enables users to complete end-to-end operations—including data import and management, data annotation, model training, model evaluation, model quantization and export, as well as testing and validation—through a no-code, visual interactive approach. The platform offers seven scenario-based modeling capabilities: classification, object detection, semantic segmentation, instance segmentation, rotated object detection, OCR, and anomaly detection, covering nearly all deep learning image inspection requirements.
SKU:
BIZ-35140
