VolumeGAN - 3D-aware Image Synthesis via Learning Structural and Textural Representations

Figure: Framework of VolumeGAN.

3D-aware Image Synthesis via Learning Structural and Textural Representations
Yinghao Xu, Sida Peng, Ceyuan Yang, Yujun Shen, Bolei Zhou
Computer Vision and Pattern Recognition (CVPR), 2022

[Paper] [Project Page] [Demo]

This paper aims at achieving high-fidelity 3D-aware images synthesis. We propose a novel framework, termed as VolumeGAN, for synthesizing images under different camera views, through explicitly learning a structural representation and a textural representation. We first learn a feature volume to represent the underlying structure, which is then converted to a feature field using a NeRF-like model. The feature field is further accumulated into a 2D feature map as the textural representation, followed by a neural renderer for appearance synthesis. Such a design enables independent control of the shape and the appearance. Extensive experiments on a wide range of datasets show that our approach achieves sufficiently higher image quality and better 3D control than the previous methods.

Usage

Setup

This repository is based on Hammer, where you can find detailed instructions on environmental setup.

Test Demo

python render.py \
    --work_dir ${WORK_DIR} \
    --checkpoint ${MODEL_PATH} \
    --num ${NUM} \
    --seed ${SEED} \
    --render_mode ${RENDER_MODE} \
    --generate_html ${SAVE_HTML} \
    volumegan-ffhq

where

• WORK_DIR refers to the path to save the results.
• MODEL_PATH refers to the path of the pretrained model, regarding which we provide
  • FFHQ-256
• NUM refers to the number of samples to synthesize.
• SEED refers to the random seed used for sampling.
• RENDER_MODE refers to the type of the rendered results, including video and shape.
• SAVE_HTML controls whether to save images as an HTML for better visualization when rendering videos.

Training

For example, users can use the following command to train VolumeGAN on FFHQ in the resolution of 256x256

./scripts/training_demos/volumegan_ffhq256.sh \
    ${NUM_GPUS} \
    ${DATA_PATH} \
    [OPTIONS]

where

• NUM_GPUS refers to the number of GPUs used for training.
• DATA_PATH refers to the path to the dataset (zip format is strongly recommended).
• [OPTIONS] refers to any additional option to pass. Detailed instructions on available options can be found via python train.py volumegan-ffhq --help.

NOTE: This demo script uses volumegan_ffhq256 as the default job_name, which is particularly used to identify experiments. Concretely, a directory with name job_name will be created under the root working directory, which is set as work_dirs/ by default. To prevent overwriting previous experiments, an exception will be raised to interrupt the training if the job_name directory has already existed. Please use --job_name=${JOB_NAME} option to specify a new job name.

Evaluation

Users can use the following command to evaluate a well-trained model

./scripts/test_metrics.sh \
    ${NUM_GPUS} \
    ${DATA_PATH} \
    ${MODEL_PATH} \
    fid \
    --G_kwargs '{"ps_kwargs":'{"perturb_mode":"none"}'}' \
    [OPTIONS]

BibTeX

@inproceedings{xu2021volumegan,
  title     = {3D-aware Image Synthesis via Learning Structural and Textural Representations},
  author    = {Xu, Yinghao and Peng, Sida and Yang, Ceyuan and Shen, Yujun and Zhou, Bolei},
  booktitle = {CVPR},
  year      = {2022}
}