Real-Time Neural Character Rendering with Pose-Guided Multiplane Images (ECCV 2022)

Project Page | Video | Paper | Dynamic MVS Data


We propose pose-guided multiplane image (MPI) synthesis which can render an animatable character in real scenes with photorealistic quality. We use a portable camera rig to capture the multi-view images along with the driving signal for the moving subject. Our method generalizes the image-to-image translation paradigm, which translates the human pose to a 3D scene representation — MPIs that can be rendered in free viewpoints, using the multi-views captures as supervision. To fully cultivate the potential of MPI, we propose depth-adaptive MPI which can be learned using variable exposure images while being robust to inaccurate camera registration. Our method demonstrates advantageous novel-view synthesis quality over the state-of-the-art approaches for characters with challenging motions. Moreover, the proposed method is generalizable to novel combinations of training poses and can be explicitly controlled. Our method achieves such expressive and animatable character rendering all in real time, serving as a promising solution for practical applications.

Getting started

Set up the conda environment:

conda env create -f environment.yml
conda activate pgmpi 

Prepare the data: Download the sample scene and unzip it in the project folder. Start training:


Use tensorboard to supervise the training:

tensorboard --logdir runs/

To evaluate or generate MVS videos:


Data preprocess

We provide a simple data preprocess scripts for the aligned videos. Extra packages needs to be installed.

pip install opencv-python== 
pip install mediapipe

To estimate the camera pose, colmap needs to be installed. Please refer to the utils/ for the estimation settings. Feel free to adjust the preprocess to your need.


Training and evaluation:

We provide several training options, including the experiment path, image resolution and so on. To enable the learnable appearance and learnable depth use the following options:


We provide different rendering settings. We can render fixed view videos or bulletpoint time videos by enable the following options:


We can also render the estimated depth image by:



We thank the authors of Nex-MPI for sharing their code.


  title={Real-Time Neural Character Rendering with Pose-Guided Multiplane Images},
  author={Ouyang, Hao and Zhang, Bo and Zhang, Pan and Yang, Hao and Yang, Jiaolong and Chen, Dong and Chen, Qifeng and Wen, Fang},
  journal={arXiv preprint arXiv:2204.11820},


View Github