Dataset of sparse 3D VR sketches

This dataset is a part of the following publication:

"Fine-Grained VR Sketching: Dataset and Insights" Ling Luo, Yulia Gryaditskaya, Yongxin Yang, Tao Xiang, Yi-Zhe Song

Proceedings of International Conference on 3D Vision (3DV) 2021.



VR Sketch Interface

The data

We present the first fine-grained dataset of 1,497 3D VR sketch and 3D shape pairs for 1,005 chair shapes with large shapes diversity from the ShapeNetCore dataset from 50 participants.

The structure of the content is stated bellow:

  • sketch_obj
    • original_obj: These are collected sketches with corresponding time information in timestamp.txt. Each sketch obj file contains a collection of strokes, where each stroke consists of several edges. Each sketch file is named as 03001627_$ShapeID_model_$PID_$timestamp_sketch.obj, where 03001627 is the ID of the chair category, $ShapeID is the shape ID in ShapeNet dataset, $PID is the participant ID, $timestamp is the creation time. Each timestamp file logs [x,y,z,t] in each line, where t is the relative time to the start the stroke when creating point [x,y,z].
    • filtered_obj: This folder contains the sketches in the OBJ file format after filtering outlier strokes. For detailed information about the filtering step, please refer to the stroke filtering script.
  • shape_obj: For completeness, we include 6,597 chair shapes from the ShapeNetCore dataset.
  • point_cloud Files containing uniformly sampled point clouds. Files are named as $ShapeID.npy. We also provide the point cloud sampling script. For sketches, the point clouds are sampled from the sketches with the filtered out outlier strokes.
    • sketch: point clouds of 1,005 sketches.
    • aligned_sketch: point clouds of 1,005 sketches aligned along X,Y,Z axes.
    • hs_id38: 544 sketches from the participant with the ID 38. This participant is used as additional training data for the experiment in the paper in Table 4: #16 Aug. style.
    • shape: point clouds of 6,597 chair shapes from the ShapeNetCore dataset.
  • image: Rendered preview images for point clouds and 3D shapes using the Mitsuba rendering script.
    • shape: Rendered preview images for point clouds of 6,597 shapes
    • sketch: Rendered preview images for point clouds of 1,005 aligned sketches
  • list
    • train.txt : name list for 702 sketch-shape pairs in train set from 45 seen participants
    • val.txt : name list for 101 sketch-shape pairs in validation set from 45 seen participants.
    • test:
      • test.txt: 202 sketches used for test.
      • test_shape.txt: 5,794 shapes used for test.
      • test_5participants.txt: 50 sketches from 5 unseen participants used for test.
      • test_45 participants.txt: 152 used from 45 seen participants used for test.

All the data: 3dv_2021_vr_sketches_full.tar.gz

VR sketches in OBJ format: 3dv_2021_vr_sketches_obj.tar.gz

Point clouds and lists: 3dv_2021_vr_sketches_train_test.tar.gz

If you use this dataset or scripts please cite:

title = {{Fine-Grained VR Sketching: Dataset and Insights}},
author = {Ling Luo, Yulia Gryaditskaya, Yongxin Yang, Tao Xiang, Yi-Zhe Song},
booktitle = {Proceedings of International Conference on 3D Vision (3DV)},
year = {2021}


For any questions about this dataset, please contact Ling Luo.