3D-CycleGANs
Reconstruction from sparse 3D models to dense 3D models by CycleGANs and extra autoencoder training.
Data and saved_models
https://drive.google.com/drive/folders/11J3flfTdlYlX7kGDFUM7MB9P1sAOv3gw?usp=sharing
Requirements
python 2.7
numpy 1.13.1
keras 2.0.8
keras-contrib 2.0.8
tensorflow-gpu 1.3.0
scipy 0.19.1
Run
$ cd cycle_gan/
$ python cyclegan.py
Validation
Generate data
$ cd Validation
$ python predict_generator.py
Intersection-over-union (IoU) Metrics
$ python IoU_test.py chair ../Data/chair/train_3d/