This repo hosts the surface region correspondence evaluation dataset consisting of mulitple surface region ground truths for 3 object categories - Chair, Car and Plane.
Each object category has 10 objects with each object having 8 different ground truth surface regions. Providing 90 pairwise correspondence estimation possibility per category, with 8 region correspondences estimations in each pair.
Sample images of surface region ground truths below:
| Chair | Car | Plane |
|---|---|---|
 |
 |
 |
NOTE1: This dataset is in association with a work titled "NRDF - Neural Region Descriptor Fields as Implicit ROI Representation for Robotic 3D Surface Processing" and is in IROS 2024 conference submission. After acceptance - Code and pretrained models for NRDF will be uploaded here -> https://github.com/Profactor/Neural-Region-Descriptor-Fields
NOTE2: This is a growing dataset and more objects categories and corresponding surface regions will be added gradually.