3DIAS_Pytorch

This repository contains the official code to reproduce the results from the paper:

3DIAS: 3D Shape Reconstruction with Implicit Algebraic Surfaces (ICCV 2021)

[project page] [arXiv]

Installation

Clone this repository into any place you want.

git clone https://github.com/myavartanoo/3DIAS_PyTorch.git
cd 3DIAS_Pytorch

Dependencies

• Python 3.8.5
• PyTorch 1.7.1
• numpy
• Pillow
• open3d
• PyMCubes (or build this repo)

Install dependencies in a conda environment.

conda create -n 3dias python=3.8
conda activate 3dias

pip install -r requirements.txt

Pretrained model

Download config.json and checkpoint-epoch#.pth from below links and save in weigths folder. Note that we get Multi-class weight by training with all-classes and Single-class weight by training with each class

Multi-class

Dropbox or Mirror

Single-class

To download all the single-class weigths, run

sh download_weights.sh

Or you can get the weights one-by-one.

airplane / bench / cabinet / car / chair / display / lamp / speaker / rifle / sofa / table / phone / vessel

Quickstart (Demo)

You can now test our demo code on the provided input images in the input folder. (Or you can use other images in shapeNet.) To this end, simply run,

CUDA_VISIBLE_DEVICES=0 python demo.py --inputimg "./input/<image_name>.png" --config "./weights/config.json" --resume "./weights/checkpoint-epoch890.pth" 

The result meshes are saved in output folder. (We've created a few example meshes)

• total.ply is a whole mesh
• parts_.ply are meshes for parts To see the mesh, you can use meshlab

If you want to visualize meshes with open3d, run with --visualize option as below.

CUDA_VISIBLE_DEVICES=0 python demo.py --inputimg "./input/<image_name>.png" --config "./weights/config.json" --resume "./weights/checkpoint-epoch890.pth" --visualize

The preprocessed dataset, training, testing code will be distributed soon.

Citation

If you find our code or paper useful, please consider citing

@inproceedings{3DIAS,
    title = {3DIAS: 3D Shape Reconstruction with Implicit Algebraic Surfaces},
    author = {Mohsen Yavartanoo, JaeYoung Chung, Reyhaneh Neshatavar, Kyoung Mu Lee},
    booktitle = {Proceedings IEEE Conf. on International Conference on Computer Vision (ICCV)},
    year = {2021}
}