Prerequisites

• Linux (tested under Ubuntu 16.04 )
• Python (tested under 3.6.2)
• TensorFlow (tested under 1.13.1-GPU )
• numpy, scipy, h5py, scipy, open3d, PyMCubes, tflearn, etc.

The code reuses some components from latent_3d_points, pointnet2 and IM-NET. Before run the code, please compile the customized TensorFlow operators under the folders "latent_3d_points/structural_losses" and "pointnet_plusplus/tf_ops".

Dataset

• Download the dataset and pretained models HERE.

Usage

The commond lines for training and testing the models are all under the folder "./CMD_sh". You may need to open, read and modify the .sh files.

To train and test part alignment:

bash ./CMD_sh/partAlign_train_chair.sh

To train and test joint synthesis:

% first pretrain part encoders
bash ./CMD_sh/partAE_train_chair1234.sh

% then train the joint synthesis network and test it on input parts with GT joints
bash ./CMD_sh/jointSynthesis_train_chair.sh

To test joint synthesis for given parts from different objects:

First set diffShape="1" in "./CMD_sh/partAlign_train_chair.sh". And run it to export aligned parts randomly selected from different objects.

Then run the test on the aligned parts:

bash ./CMD_sh/jointSynthesis_test_onRealOutput_chair.sh

Poisson blending

Take a look at the folder "poisson-blending"

Point samping and data preprocessing

Take a look at the folder "data-preprocess"

Citation

If you find our work useful in your research, please consider citing:

@inproceedings{yin2020coalesce,
    author = {Kangxue Yin, Zhiqin Chen, Siddhartha Chaudhuri, Matthew Fisher, Vladimir Kim and Hao Zhang}
    title = {COALESCE: Component Assembly by Learning to Synthesize Connections}
    booktitle = {Proc. of 3DV}
    year = {2020}
}