artonson / def

Official implementation of the SIGGRAPH-2022 paper "DEF: Deep Estimation of Sharp Geometric Features in 3D Shapes"

DEF: Deep estimation of sharp geometric features in 3D shapes

SIGGRAPH 2022 [Project Page] [Arxiv] [Bibtex]

This is an official implementation of the paper Albert Matveev, Ruslan Rakhimov, Alexey Artemov, Gleb Bobrovskikh, Vage Egiazarian, Emil Bogomolov, Daniele Panozzo, Denis Zorin, and Evgeny Burnaev. "DEF: Deep estimation of sharp geometric features in 3D shapes". ACM Trans. Graph. 41, 4, Article 108 (July 2022), 22 pages.

Getting started

Below, we enumerate the major steps required for our method to work, and provide the links to the respective documentation and resources. To get familiar with more details of how our method works, please refer to the respective documentation pages, the source code, contact the authors via [artonson at yandex ru], or open an issue.

• System installation and configuration
• Generating synthetic training datasets
• Generating real-world training datasets
• Training point-based and image-based DEF networks
• Running inference with a pre-trained DEF network
• Running reconstruction on complete 3D models
• Running extraction of parametric curves
• Obtaining predictions made by DEF

Pre-trained models

We provide a variety of pre-trained DEF networks (both image-based and point-based). See Training networks page for downloading the respective weight files.

Training and evaluation datasets

See Synthetic data and Real-world data pages.

Citing

@article{10.1145/3528223.3530140,
author = {Matveev, Albert and Rakhimov, Ruslan and Artemov, Alexey and Bobrovskikh, Gleb and Egiazarian, Vage and Bogomolov, Emil and Panozzo, Daniele and Zorin, Denis and Burnaev, Evgeny},
title = {DEF: Deep Estimation of Sharp Geometric Features in 3D Shapes},
year = {2022},
issue_date = {July 2022},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
volume = {41},
number = {4},
issn = {0730-0301},
url = {https://doi.org/10.1145/3528223.3530140},
doi = {10.1145/3528223.3530140},
journal = {ACM Trans. Graph.},
month = {jul},
articleno = {108},
numpages = {22},
keywords = {curve extraction, sharp geometric features, deep learning}
}

Acknowledgements

We are grateful to Prof. Dzmitry Tsetserukou (Skoltech) and his laboratory staff for providing the 3D printing device and technical support. We thank Sebastian Koch (Technical University of Berlin), Timofey Glukhikh (Skoltech) and Teseo Schneider (New York University) for providing assistance in data generation. We also thank Maria Taktasheva (Skoltech) for assistance in computational experiments. We acknowledge the use of computational resources of the Skoltech CDISE supercomputer Zhores for obtaining the results presented in this paper. The work was supported by the Analytical center under the RF Government (subsidy agreement 000000D730321P5Q0002, Grant No. 70-2021-00145 02.11.2021).

def's Issues

ValueError: too many values to unpack (expected 2)

There are some promblem in the generating synthetic training datasets
When I try to follow the Example data generation
I got error

09.05.2023 03:24:46.804851  ForkPoolWorker-1 Computing images from pose 0
09.05.2023 03:24:46.805265  ForkPoolWorker-1   getting image
09.05.2023 03:24:47.432734  ForkPoolWorker-1   done
09.05.2023 03:24:47.432837  ForkPoolWorker-1   processing features
09.05.2023 03:24:48.483014  ForkPoolWorker-1   done
09.05.2023 03:24:48.485330  ForkPoolWorker-1 Error processing item 00000002_1ffb81a71e5b402e966b9341_001 from chunk [/data/abc_0000_obj_v00.7z,/data/abc_0000_feat_v00.7z]: too many values to unpack (expected 2)
09.05.2023 03:24:48.488249  ForkPoolWorker-1 Traceback (most recent call last):
  File "/code/scripts/data_scripts/generate_depthmap_data.py", line 215, in generate_patches_abc
    for configuration, patch_info in get_annotated_patches(mesh, features, item.item_id, config):
  File "/code/scripts/data_scripts/generate_depthmap_data.py", line 146, in get_annotated_patches
    z_direction=np.array([0., 0., -1.])):
ValueError: too many values to unpack (expected 2)

Link error in "Generating synthetic training datasets" section

The link about [synthetic_data/complete_models/] in "Generating synthetic training datasets" section is redirecting on the patch Dropbox and not the complete models.

About releasing Dataset and the processing code

Hello, great work!
Do you have any schedule for releasing the Synthetic Datasets "DEF-Sim" and its corresponding processing code for the ABC dataset?