ouasfi/fs-sdf

[ECCV'22] "Few 'Zero Level Set'-Shot Learning of Shape Signed Distance Functions in Feature Space"

Few 'Zero Level Set'-Shot Learning of Shape Signed Distance Functions in Feature Space

Amine Ouasfi, Adnane Boukhayma.
ECCV 2022

Install

Please follow the instructions detailed in IF-Net. Then, install the required packages for this project by running:

pip install torch==1.9.1+cu111 torchvision==0.10.1+cu111 torchaudio==0.9.1 -f https://download.pytorch.org/whl/torch_stable.html

Demo (New)

To run our pretrained model on your own shapes, place them in a directory and create a .lst file containing your file names. Then run ,

python demo.py -data_path [path to your data directory]  -exp <exp_name> -checkpoint <checkpoint> -inner_steps 5  -pc_samples 3000 -res 128 -batch_size 8 -save_path gen_dir/

where res is the resolution of the encoder's grid and exp_name is the path to the folder containing the trained model checkpoints.

Data Preparation

The data preparation process is the same as in IF-Net.

To generate ground truth Signed distance values instead of occupancies run:

python data_processing/boundary_sampling_sdf.py -sigma 0.1
python data_processing/boundary_sampling_sdf.py -sigma 0.01

Training

To start a training please run the following command:

python train.py -res 128 -pc_samples 3000 -epochs 100 -inner_steps 5 -batch_size 8

You can add the following options -p_enc and _p_dec to initialize the encoder and/or the decoder with a pretrained model. Also, you can freeze the encoder during the training by adding the option -freeze to your command.

Generation

The command:

python generate.py -res 128 -pc_samples 3000 -batch_size 8 -inner_steps 5 -exp <exp_name> -checkpoint <checkpoint>  

Where exp_name is the path to the folder containing the trained model checkpoints.

Evaluation

The evaluation process is the same as in IF-Net.

Pretrained Models

Pretrained models can be found here.

Contact

For questions and comments regarding the code please contact Amine Ouasfi via mail. (See Paper)