Skip to content

Dataset · UMTRI

Synthetic data for material identification and properties from photos

UMTRI/MatPredict

Paired RGB object renders, material property maps, and segmentation masks for inverse rendering tasks.

What's inside

Contains synthetic renders of various object variants, each accompanied by physical material property maps (albedo, roughness, metallic), region and material class segmentation masks, camera matrices, and metadata. Language and exact sample count are not specified in the card.

What you can do with it

  • Training inverse rendering models to reconstruct PBR maps (albedo, roughness, metallic) from a single RGB image
  • Training neural networks for semantic material segmentation in images
  • Creating 3D assets and transferring textures in computer graphics pipelines
Tasks
Image → image, Image segmentation
Downloads this month
77.9K
Likes
1

How to load it

from datasets import load_dataset

ds = load_dataset("UMTRI/MatPredict")