This is the code project for Fine-Grained VR Sketching: Dataset and Insights published on 3DV 2021.
Paper Link: [Paper] [Supplemental]
SkethchyVR dataset are avaiable at Dataset webpage and Google Drive (point cloud only).
Here are some samples of the shape sketch pairs in SkethchyVR:
The dataset used in this project is collected with a VR sketching interface called SketchyVR which allow participants to sketch inmmersively using VR headsets and handles like Oculus Rift.
Demonstration on filtering the original sketches: tools/Filter original sketch.ipynb
Once filtering the original VR sketches, point clouds for training are sampled form the filtered sketch. Script for sampling from sketches and meshes: tools/gen_pointcloud.py
Render image for point cloud files: tools/vis_pc_mitsuba.py
Install MITSUBA first and then replace the PATH_TO_MITSUBA2 with your path.
Train 3D sketch based 3D shape retrieval:
train_triplet_3dv.py
Train 2D sketch based 3D shape retrieval:
train_triplet_view_2d.py
The val.txt in the published dataset only includes 101 chair models from ShapeNetCore. To make the validation more reliable, I added chair shapes from ModelNet10, resulting in a new list file, val_shape.txt. Therefore, val_shape.txt contains many more chair model names from ModelNet10 compared to val.txt. You can choose which one to use based on your needs. All the list files, including val_shape.txt can be found here. Additionally, the required chair shapes from ModelNet10 can be downloaded from this link.
Please cite our work if you find it useful:
@inproceedings{luo2021fine,
title={Fine-Grained VR Sketching: Dataset and Insights.},
author={Luo, Ling and Gryaditskaya, Yulia and Yang, Yongxin and Xiang, Tao and Song, Yi-Zhe},
booktitle={2021 International Conference on 3D Vision (3DV)},
pages={1003--1013},
year={2021},
organization={IEEE}
}