Please follow DeepSDF for environment installation.

• Please follow DeepSDF for preprocess the shapes and get SDF samples (both SdfSamples and SurfaceSamples).
• Use python process_data.py to assign each SDF point sample to a local shape.
• Use python process_data_normal.py to pre-compute coordinate field initialization for each local shape.

• Use ./train.sh for training
• Use ./reconstruct.sh for testing. Use ./reconstruct_batch.py for fast validation.

The trained MLP can be found here

This repo is developed from DeepSDF and DeepLS