EXIM: A Hybrid Explicit-Implicit Representation for Text-Guided 3D Shape Generation

SIGGRAPH Asia 2023 & ACM TOG

Project Page

Code for the paper EXIM: A Hybrid Explicit-Implicit Representation for Text-Guided 3D Shape Generation.

Authors: Zhengzhe Liu, Jingyu Hu, Ka-Hei Hui, Xiaojuan Qi, Daniel Cohen-Or, Chi-Wing Fu

Installation

conda env create -f environment.yaml
conda activate EXIM
cd stage 2
python setup.py build_ext --inplace
cd data_processing/libmesh/
python setup.py build_ext --inplace
cd ../libvoxelize/
python setup.py build_ext --inplace
cd ../..

Data Preparation

Download our models and list of files.

unzip to "EXIM/data/"

3D Shape Generation

cd stage1
python test_chair.py
cd ../stage2
sh test.sh

Training

(1) Stage 1

Download the train data.

Put to "EXIM/data/"

cd stage1
python trainer_new.py

(2) Stage 2

Download Choy et. al. rendering data and IF-Net data: 03001627.tar.gz

Put to "EXIM/data/"

cd stage2
sh train.sh

Evaluation

cd evaluation

Manipulation

Option1: Locate the interested region following Diff-Edit.

cd manipulation

python mani_diffedit.py

Option2: Locate the interested region using Interactive System (Thanks to Ruihui Li, Ka-Hei Hui, and Jingyu Hu for developing this tool).

First, run the UI-Interface

python sample_point_cloud.py
cd UI-Interface
python label_interface.py

Load a input.obj.ply Select a region you want to edit The "selection.npy" file is saved in "UI-Interface/debug/" Move the selection.npy to the "manipulation folder"

python mani_select.py

To train the manipulation model:

python seg.py
python trainer_new.py

Acknowledgement

The code is built upon Wavelet-Diffusion and DVR

Contact

If you have any questions or suggestions about this repo, please feel free to contact me (liuzhengzhelzz@gmail.com).

liuzhengzhe/EXIM