We present NeX, a new approach to novel view synthesis based on enhancements of multiplane image (MPI) that can reproduce NeXt-level view-dependent effects---in real time. Unlike traditional MPI that uses a set of simple RGBα planes, our technique models view-dependent effects by instead parameterizing each pixel as a linear combination of basis functions learned from a neural network. Moreover, we propose a hybrid implicit-explicit modeling strategy that improves upon fine detail and produces state-of-the-art results. Our method is evaluated on benchmark forward-facing datasets as well as our newly-introduced dataset designed to test the limit of view-dependent modeling with significantly more challenging effects such as the rainbow reflections on a CD. Our method achieves the best overall scores across all major metrics on these datasets with more than 1000× faster rendering time than the state of the art.

Getting started

conda env create -f environment.yml
conda activate nex
python -scene data/crest_demo -model_dir crest -http
tensorboard --logdir runs/


We provide environment.yml to help you setup a conda environment.

conda env create -f environment.yml


Shiny dataset

Download: Shiny dataset.

We provide 2 directories named shiny and shiny_extended.

  • shiny contains benchmark scenes used to report the scores in our paper.
  • shiny_extended contains additional challenging scenes used on our website project page and video

NeRF's real forward-facing dataset

Download: Undistorted front facing dataset

For real forward-facing dataset, NeRF is trained with the raw images, which may contain lens distortion. But we use the undistorted images provided by COLMAP.

However, you can try running other scenes from Local lightfield fusion (Eg. airplant) without any changes in the dataset files. In this case, the images are not automatically undistorted.

Deepview's spaces dataset

Download: Modified spaces dataset

We slightly modified the file structure of Spaces dataset in order to determine the plane placement and split train/test sets.

Using your own images.

Running NeX on your own images. You need to install COLMAP on your machine.

Then, put your images into a directory following this structure

|-- images
     | -- image_name1.jpg
     | -- image_name2.jpg

The training code will automatically prepare a scene for you. You may have to tune planes.txt to get better reconstruction (see dataset explaination)


Run with the paper's config

python -scene ${PATH_TO_SCENE} -model_dir ${MODEL_TO_SAVE_CHECKPOINT} -http

This implementation uses scikit-image to resize images during training by default. The results and scores in the paper are generated using OpenCV's resize function. If you want the same behavior, please add -cv2resize argument.

Note that this code is tested on an Nvidia V100 32GB and 4x RTX 2080Ti GPU.

For a GPU/GPUs with less memory (e.g., a single RTX 2080Ti), you can run using the following command:

python -scene ${PATH_TO_SCENE} -model_dir ${MODEL_TO_SAVE_CHECKPOINT} -http -layers 12 -sublayers 6 -hidden 256

Note that when your GPU runs ouut of memeory, you can try reducing the number of layers, sublayers, and sampled rays.


To generate a WebGL viewer and a video result.

python -scene ${scene} -model_dir ${MODEL_TO_SAVE_CHECKPOINT} -predict -http

Video rendering

To generate a video that matches the real forward-facing rendering path, add -nice_llff argument, or -nice_shiny for shiny dataset


    author = {Wizadwongsa, Suttisak and Phongthawee, Pakkapon and Yenphraphai, Jiraphon and Suwajanakorn, Supasorn},
    title = {NeX: Real-time View Synthesis with Neural Basis Expansion},
    booktitle = {IEEE Conference on Computer Vision and Pattern Recognition (CVPR)}, 
    year = {2021},