PASMVS : a perfectly accurate, synthetic, path-traced dataset featuring specular material properties for multi-view stereopsis training and reconstruction applications

Abstract

A Perfectly Accurate, Synthetic dataset for Multi-View Stere- opsis (PASMVS) is presented, consisting of 400 scenes and 18,0 0 0 model renderings together with ground truth depth maps, camera intrinsic and extrinsic parameters, and binary segmentation masks. Every scene is rendered from 45 differ- ent camera views in a circular pattern, using Blender’s path- tracing rendering engine. Every scene is composed from a unique combination of two camera focal lengths, four 3D models of varying geometrical complexity, five high defini- tion, high dynamic range (HDR) environmental textures to replicate photorealistic lighting conditions and ten materials. The material properties are primarily specular, with a selec- tion of more diffuse materials for reference. The combination of highly specular and diffuse material properties increases the reconstruction ambiguity and complexity for MVS recon- struction algorithms and pipelines, and more recently, state- of-the-art architectures based on neural network implemen- tations. PASMVS serves as an addition to the wide spectrum of available image datasets employed in computer vision re- search, improving the precision required for novel research applications.