• DirectVoxGO
• • • Custom casual capturing
    • Features
    • Installation
  • Directory structure for the datasets
  • GO
  • • Custom casually captured scenes
    • Development and tuning guide
    • • Extention to new dataset
      • Speed and quality tradeoff
  • Advanced data structure
  • Acknowledgement

DirectVoxGO

Direct Voxel Grid Optimization (CVPR2022 Oral, project page, DVGO paper, DVGO v2 paper).

https://user-images.githubusercontent.com/2712505/153380311-19d6c3a1-9130-489a-af16-ad36c78f10a9.mp4

https://user-images.githubusercontent.com/2712505/153380197-991d1689-6418-499c-a192-d757f9a64b64.mp4

Custom casual capturing

A short guide to capture custom forward-facing scenes and rendering fly-through videos.

Below are two rgb and depth fly-through videos from custom captured scenes.

https://user-images.githubusercontent.com/2712505/174267754-619d4f81-dd04-4c50-ba7f-434774cb890e.mp4

Features

• Speedup NeRF by replacing the MLP with the voxel grid.
• Simple scene representation:
  • Volume densities: dense voxel grid (3D).
  • View-dependent colors: dense feature grid (4D) + shallow MLP.
• Pytorch cuda extention built just-in-time for another 2--3x speedup.
• O(N) realization for the distortion loss proposed by mip-nerf 360.
  • The loss improves our training time and quality.
  • We have released a self-contained pytorch package: torch_efficient_distloss.
  • Consider a batch of 8192 rays X 256 points.
    • GPU memory consumption: 6192MB => 96MB.
    • Run times for 100 iters: 20 sec => 0.2sec.
• Supported datasets:
  • Bounded inward-facing: NeRF, NSVF, BlendedMVS, T&T (masked), DeepVoxels.
  • Unbounded inward-facing: T&T, LF, mip-NeRF360.
  • Foward-facing: LLFF.

Installation

git clone git@github.com:sunset1995/DirectVoxGO.git
cd DirectVoxGO
pip install -r requirements.txt

Pytorch and torch_scatter installation is machine dependent, please install the correct version for your machine.

Dependencies (click to expand)

• PyTorch, numpy, torch_scatter: main computation.
• scipy, lpips: SSIM and LPIPS evaluation.
• tqdm: progress bar.
• mmcv: config system.
• opencv-python: image processing.
• imageio, imageio-ffmpeg: images and videos I/O.
• Ninja: to build the newly implemented torch extention just-in-time.
• einops: torch tensor shaping with pretty api.
• torch_efficient_distloss: O(N) realization for the distortion loss.

Directory structure for the datasets

(click to expand;)

data
├── nerf_synthetic     # Link: https://drive.google.com/drive/folders/128yBriW1IG_3NJ5Rp7APSTZsJqdJdfc1
│   └── [chair|drums|ficus|hotdog|lego|materials|mic|ship]
│       ├── [train|val|test]
│       │   └── r_*.png
│       └── transforms_[train|val|test].json
│
├── Synthetic_NSVF     # Link: https://dl.fbaipublicfiles.com/nsvf/dataset/Synthetic_NSVF.zip
│   └── [Bike|Lifestyle|Palace|Robot|Spaceship|Steamtrain|Toad|Wineholder]
│       ├── intrinsics.txt
│       ├── rgb
│       │   └── [0_train|1_val|2_test]_*.png
│       └── pose
│           └── [0_train|1_val|2_test]_*.txt
│
├── BlendedMVS         # Link: https://dl.fbaipublicfiles.com/nsvf/dataset/BlendedMVS.zip
│   └── [Character|Fountain|Jade|Statues]
│       ├── intrinsics.txt
│       ├── rgb
│       │   └── [0|1|2]_*.png
│       └── pose
│           └── [0|1|2]_*.txt
│
├── TanksAndTemple     # Link: https://dl.fbaipublicfiles.com/nsvf/dataset/TanksAndTemple.zip
│   └── [Barn|Caterpillar|Family|Ignatius|Truck]
│       ├── intrinsics.txt
│       ├── rgb
│       │   └── [0|1|2]_*.png
│       └── pose
│           └── [0|1|2]_*.txt
│
├── deepvoxels         # Link: https://drive.google.com/drive/folders/1ScsRlnzy9Bd_n-xw83SP-0t548v63mPH
│   └── [train|validation|test]
│       └── [armchair|cube|greek|vase]
│           ├── intrinsics.txt
│           ├── rgb/*.png
│           └── pose/*.txt
│
├── nerf_llff_data     # Link: https://drive.google.com/drive/folders/128yBriW1IG_3NJ5Rp7APSTZsJqdJdfc1
│   └── [fern|flower|fortress|horns|leaves|orchids|room|trex]
│
├── tanks_and_temples  # Link: https://drive.google.com/file/d/11KRfN91W1AxAW6lOFs4EeYDbeoQZCi87/view?usp=sharing
│   └── [tat_intermediate_M60|tat_intermediate_Playground|tat_intermediate_Train|tat_training_Truck]
│       └── [train|test]
│           ├── intrinsics/*txt
│           ├── pose/*txt
│           └── rgb/*jpg
│
├── lf_data            # Link: https://drive.google.com/file/d/1gsjDjkbTh4GAR9fFqlIDZ__qR9NYTURQ/view?usp=sharing
│   └── [africa|basket|ship|statue|torch]
│       └── [train|test]
│           ├── intrinsics/*txt
│           ├── pose/*txt
│           └── rgb/*jpg
│
├── 360_v2             # Link: https://jonbarron.info/mipnerf360/
│   └── [bicycle|bonsai|counter|garden|kitchen|room|stump]
│       ├── poses_bounds.npy
│       └── [images_2|images_4]
│
├── nerf_llff_data     # Link: https://drive.google.com/drive/folders/14boI-o5hGO9srnWaaogTU5_ji7wkX2S7
│   └── [fern|flower|fortress|horns|leaves|orchids|room|trex]
│       ├── poses_bounds.npy
│       └── [images_2|images_4]
│
└── co3d               # Link: https://github.com/facebookresearch/co3d
    └── [donut|teddybear|umbrella|...]
        ├── frame_annotations.jgz
        ├── set_lists.json
        └── [129_14950_29917|189_20376_35616|...]
            ├── images
            │   └── frame*.jpg
            └── masks
                └── frame*.png

GO

• Training

  $ python run.py --config configs/nerf/lego.py --render_test
  

  Use --i_print and --i_weights to change the log interval.

• Evaluation
  To only evaluate the testset PSNR, SSIM, and LPIPS of the trained lego without re-training, run:

  $ python run.py --config configs/nerf/lego.py --render_only --render_test \
                                                --eval_ssim --eval_lpips_vgg
  

  Use --eval_lpips_alex to evaluate LPIPS with pre-trained Alex net instead of VGG net.

• Render video

  $ python run.py --config configs/nerf/lego.py --render_only --render_video
  

  Use --render_video_factor 4 for a fast preview.

• Reproduction: all config files to reproduce our results.

  (click to expand)

    $ ls configs/*
    configs/blendedmvs:
    Character.py  Fountain.py  Jade.py  Statues.py
  
    configs/nerf:
    chair.py  drums.py  ficus.py  hotdog.py  lego.py  materials.py  mic.py  ship.py
  
    configs/nsvf:
    Bike.py  Lifestyle.py  Palace.py  Robot.py  Spaceship.py  Steamtrain.py  Toad.py  Wineholder.py
  
    configs/tankstemple:
    Barn.py  Caterpillar.py  Family.py  Ignatius.py  Truck.py
  
    configs/deepvoxels:
    armchair.py  cube.py  greek.py  vase.py
  
    configs/tankstemple_unbounded:
    M60.py  Playground.py  Train.py  Truck.py
  
    configs/lf:
    africa.py  basket.py  ship.py  statue.py  torch.py
  
    configs/nerf_unbounded:
    bicycle.py  bonsai.py  counter.py  garden.py  kitchen.py  room.py  stump.py
  
    configs/llff:
    fern.py  flower.py  fortress.py  horns.py  leaves.py  orchids.py  room.py  trex.py
  

Custom casually captured scenes

Coming soon hopefully.

Development and tuning guide

Extention to new dataset

Adjusting the data related config fields to fit your camera coordinate system is recommend before implementing a new one.
We provide two visualization tools for debugging.

1. Inspect the camera and the allocated BBox.
   • Export via --export_bbox_and_cams_only {filename}.npz:

     python run.py --config configs/nerf/mic.py --export_bbox_and_cams_only cam_mic.npz
     

   • Visualize the result:

     python tools/vis_train.py cam_mic.npz
     

2. Inspect the learned geometry after coarse optimization.
   • Export via --export_coarse_only {filename}.npz (assumed coarse_last.tar available in the train log):

     python run.py --config configs/nerf/mic.py --export_coarse_only coarse_mic.npz
     

   • Visualize the result:

     python tools/vis_volume.py coarse_mic.npz 0.001 --cam cam_mic.npz
     

Inspecting the cameras & BBox	Inspecting the learned coarse volume

Speed and quality tradeoff

We have reported some ablation experiments in our paper supplementary material.
Setting N_iters, N_rand, num_voxels, rgbnet_depth, rgbnet_width to larger values or setting stepsize to smaller values typically leads to better quality but need more computation.
The weight_distortion affects the training speed and quality as well.
Only stepsize is tunable in testing phase, while all the other fields should remain the same as training.

Advanced data structure

• Octree — Plenoxels: Radiance Fields without Neural Networks.
• Hash — Instant Neural Graphics Primitives with a Multiresolution Hash Encoding.
• Factorized components — TensoRF: Tensorial Radiance Fields.

You will need them for scaling to a higher grid resolution. But we believe our simplest dense grid could still be your good starting point if you have other challenging problems to deal with.

Acknowledgement

The code base is origined from an awesome nerf-pytorch implementation, but it becomes very different from the code base now.