VIVID - Novel View Synthesis with Pixel-Space Diffusion Models
_{Official PyTorch implementation}

Novel View Synthesis with Pixel-Space Diffusion Models
Noam Elata, Bahjat Kawar, Yaron Ostrovsky-Berman, Miriam Farber, Ron Sokolovsky
https://arxiv.org/abs/2411.07765

Abstract: Synthesizing a novel view from a single input image is a challenging task. Traditionally, this task was approached by estimating scene depth, warping, and inpainting, with machine learning models enabling parts of the pipeline. More recently, generative models are being increasingly employed in novel view synthesis (NVS), often encompassing the entire end-to-end system. In this work, we adapt a modern diffusion model architecture for end-to-end NVS in the pixel space, substantially outperforming previous state-of-the-art (SOTA) techniques. We explore different ways to encode geometric information into the network. Our experiments show that while these methods may enhance performance, their impact is minor compared to utilizing improved generative models. Moreover, we introduce a novel NVS training scheme that utilizes single-view datasets, capitalizing on their relative abundance compared to their multi-view counterparts. This leads to improved generalization capabilities to scenes with out-of-domain content.

Requirements and Setup

Environment

Please install a pytorch > 2 environment on a cuda-supported system, and add the following additional python packages:

conda install kornia diffusers transformers accelerate tqdm click Pillow psutil requests scipy tensorboard einops numpy scikit-image

For using depth-based models, please also install DepthAnythingV2.

pip install git+https://github.com/DepthAnything/Depth-Anything-V2

Dataset Download

RealEstate10K: Download the .txt files from the original published source here. We follow the download script provided by Learning to Render Novel Views from Wide-Baseline Stereo Pairs for downloading RealEstate10K.

Single Image Dataset: Download OpenImagesV5 using the instructions here or any alternative image dataset.

Please place the files in the following format:

> ./data
data
├── RealEstate10K
│  ├── test
│  │   ├── 000c3ab189999a83.txt
│  │   └──...
│  └── train
│      ├── 0000cc6d8b108390.txt
│      └──...
├── test
│   ├── 000c3ab189999a83
│   │   ├── 45979267.png
│   │   └──...
│   └── ...
├── train
│   │── 0000cc6d8b108390
│   │   ├── 30874000.png
│   │   └──...
│   └── ...
└── SingleImages
    └── 6134661.png

Sampling from VIVID

To sample from out model, please run the following:

python generate_images.py --preset=vivid --outdir=out

The above command automatically downloads the necessary models.

# Expanded command line for --preset=vivid
python generate_images.py \
    --net=https://ml-site.cdn-apple.com/models/vivid/vivid-base.pkl \
    --gnet=https://ml-site.cdn-apple.com/models/vivid/vivid-uncond.pkl \
    --sr-model=https://ml-site.cdn-apple.com/models/vivid/vivid-sr.pkl \
    --guidance=1.5 \
    --range-selection mid \
    --outdir=out

Calculating metrics

To calculate metrics (FID and PSNR, etc.) directly, please run the following script:

torchrun --standalone --nproc_per_node=8 calculate_metrics.py gen --num 10000 \
    --net=https://ml-site.cdn-apple.com/models/vivid/vivid-base.pkl \
    --gnet=https://ml-site.cdn-apple.com/models/vivid/vivid-uncond.pkl \
    --sr-model=https://ml-site.cdn-apple.com/models/vivid/vivid-sr.pkl \
    --guidance=1.5 \
    --range-selection mid

Please use calculate_metrics.py gen --help to see a full list of possible argument.

Having generated images with the generate_images.py, metrics can be computed using:

python calculate_metrics.py calc --images=out

We also provide programatic APIs similar for computing metrics.

Training new models

New models can be trained using train_nvs.py. To reproduce our results, run:

torchrun --standalone --nproc_per_node=8 train_nvs.py --preset=vivid-base 
torchrun --standalone --nproc_per_node=8 train_nvs.py --preset=vivid-sr 
torchrun --standalone --nproc_per_node=8 train_nvs.py --preset=vivid-uncond 

See python train_nvs.py --help for the full list of options.

License

The code in this repository is released under the terms detailed in LICENSE.txt, and uses code originally distributed under the license detaled in ACKNOWLEDGEMENTS.txt. The accompanying model weights are released under the license listed in LICENSE-MODEL.txt.

Citation

@article{elata2024novel,
  title     = {Novel View Synthesis with Pixel-Space Diffusion Models},
  author    = {Noam Elata and Bahjat Kawar and Yaron Ostrovsky-Berman and Miriam Farber and Ron Sokolovsky},
  journal   = {arXiv preprint arXiv:2411.07765},
  year      = {2024}
}

Acknowledgments

This code is based on the original EDM2 implementation, please consider also citing:

@inproceedings{Karras2024edm2,
  title     = {Analyzing and Improving the Training Dynamics of Diffusion Models},
  author    = {Tero Karras and Miika Aittala and Jaakko Lehtinen and
               Janne Hellsten and Timo Aila and Samuli Laine},
  booktitle = {Proc. CVPR},
  year      = {2024},
}