Stereo Magnification: Learning view synthesis using multiplane images

This repository reproduces Pytorch implementation of "Stereo Magnification: Learning View Synthesis using Multiplane Images"

The code is based on authors' Tensorflow implementation here and reproduce implementation here. Thanks for sharing the codes.

example images with the original tensorflow weight, which I have converted to PyTorch state dicts.

Datasets

1. Download real-estate-10k-dataset.
2. Set your dataset path like this:

$/real-estate-10k
├──── Training_set
│      ├──── train
│      └──── test

3. Run python downloadAndProcess.py (youtube-dl needed).

Train

Run the following command to train Stereo Magnification:

python train.py

Test

Run the following command to evaluate Stereo Magnification:

python test.py

Demo

Run the following command to make MPIs viewer:

python demo.py