Deep Implicit Surface Point Prediction Networks

Project Page | Paper

If you find our code or paper useful, please cite as

@InProceedings{Venkatesh_2021_ICCV,
    author    = {Venkatesh, Rahul and Karmali, Tejan and Sharma, Sarthak and Ghosh, Aurobrata and Babu, R. Venkatesh and Jeni, Laszlo A. and Singh, Maneesh},
    title     = {Deep Implicit Surface Point Prediction Networks},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2021},
    pages     = {12653-12662}
}

Setting up environment

Create an anaconda environment called mesh_funcspace using

conda env create -f environment.yml
conda activate mesh_funcspace

Single Shape CSP

Training

To train a on a new 3D shape, run

python single_shape_csp/pt_pred/train.py -ename=EXP_NAME -infile=OFF_FILE_PATH

for eg.

python single_shape_csp/pt_pred/train.py -ename=lion -infile=./single_shape_csp/data/lion.off

Evaluation

For evaluation of the models, run:

python single_shape_csp/pt_pred/test.py -ename=EXP_NAME -model_iter=MODEL_ITER -reverse -num_views=NUM_VIEWS

where EXP_NAME is the experiment name to pick the weights from, MODEL_ITER is the checkpoint in EXP NAME (which can can be found in single_shape_csp/weights/EXP_NAME/MODEL_ITER.pt), and NUM_VIEWS is number of azimuths to uniformly sample for multi-view rendering.

The generated files can be found at single_shape_csp/videos/EXP_NAME.

Multi-shape CSP

Multi-shape CSP code can be found in surface_recon.