torch-ngp

A pytorch implementation of instant-ngp, as described in Instant Neural Graphics Primitives with a Multiresolution Hash Encoding.

News: With the CUDA ray marching option for NeRF, we can:

• converge to a reasonable result in ~1min (50 epochs).
• render a 1920x1080 image in ~1s.

For the LEGO dataset, we can reach ~20FPS at 800x800 due to efficient voxel pruning.

(Tested on the fox dataset with a TITAN RTX. The speed is still 2-5x slower compared to the original implementation.)

A GUI for training/visualizing NeRF is also available!

Progress

As the official pytorch extension tinycudann has been released, the following implementations can be used as modular alternatives. The performance and speed of these modules are guaranteed to be on-par, and we support using tinycudann as the backbone by the --tcnn flag. Later development will be focused on reproducing the NeRF inference speed.

• Fully-fused MLP
  • basic pytorch binding of the original implementation
• HashGrid Encoder
  • basic pytorch CUDA extension
  • fp16 support
• Experiments
  • SDF
    • baseline
    • better SDF calculation (especially for non-watertight meshes)
  • NeRF
    • baseline
    • ray marching in CUDA.
• NeRF GUI
  • supports training.
• Misc.
  • improve rendering quality of cuda raymarching!
  • improve speed (e.g., avoid the cat in NeRF forward)
  • support visualize/supervise normals (add rendering mode option).
  • support blender dataset format.

Install

pip install -r requirements.txt

# (optional) install the tcnn backbone
pip install git+https://github.com/NVlabs/tiny-cuda-nn/#subdirectory=bindings/torch

Tested on Ubuntu with torch 1.10 & CUDA 11.3 on TITAN RTX.

Usage

We use the same data format as instant-ngp, e.g., armadillo and fox. Please download and put them under ./data.

First time running will take some time to compile the CUDA extensions.

# train with different backbones (with slower pytorch ray marching)
# for the colmap dataset, the default dataset setting `--mode colmap --bound 2 --scale 0.33` is used.
python main_nerf.py data/fox --workspace trial_nerf # fp32 mode
python main_nerf.py data/fox --workspace trial_nerf --fp16 # fp16 mode (pytorch amp)
python main_nerf.py data/fox --workspace trial_nerf --fp16 --ff # fp16 mode + FFMLP (this repo's implementation)
python main_nerf.py data/fox --workspace trial_nerf --fp16 --tcnn # fp16 mode + official tinycudann's encoder & MLP

# test mode
python main_nerf.py data/fox --workspace trial_nerf --fp16 --ff --test

# use CUDA to accelerate ray marching (much more faster!)
python main_nerf.py data/fox --workspace trial_nerf --fp16 --ff --cuda_ray # fp16 mode + FFMLP + cuda raymarching

# start a GUI for NeRF training & visualization
# always use with `--fp16 --ff/tcnn --cuda_ray` for an acceptable framerate!
python main_nerf.py data/fox --workspace trial_nerf --fp16 --ff --cuda_ray --gui

# test mode for GUI
python main_nerf.py data/fox --workspace trial_nerf --fp16 --ff --cuda_ray --gui --test

# for the blender dataset, you should add `--mode blender --bound 1 --scale 0.8`
# --mode specifies dataset type ('blender' or 'colmap')
# --bound means the scene is assumed to be inside box[-bound, bound]
# --scale adjusts the camera locaction to make sure it falls inside the above bounding box.
python main_nerf.py data/nerf_synthetic/lego --workspace trial_nerf --fp16 --ff --cuda_ray --mode blender --bound 1 --scale 0.8 
python main_nerf.py data/nerf_synthetic/lego --workspace trial_nerf --fp16 --ff --cuda_ray --mode blender --bound 1 --scale 0.8 --gui

check the scripts directory for more provided examples.

Difference from the original implementation

• Instead of assuming the scene is bounded in the unit box [0, 1] and centered at (0.5, 0.5, 0.5), this repo assumes the scene is bounded in box [-bound, bound], and centered at (0, 0, 0). Therefore, the functionality of aabb_scale is replaced by bound here.
• For the hashgrid encoder, this repo only implement the linear interpolation mode.
• For the voxel pruning in ray marching kernels, this repo doesn't implement the multi-scale density grid (check the mip keyword), and only use one 128x128x128 grid for simplicity. Instead of updating the grid every 16 steps, we update it every epoch, which may lead to slower first few epochs if using --cuda_ray.
• For the blender dataest, the default mode in instant-ngp is to load all data (train/val/test) for training. Instead, we only use the specified split to train in CMD mode for easy evaluation. However, for GUI mode, we follow instant-ngp and use all data to train (check type='all' for NeRFDataset).

Update Logs

• 3.1: add type='all' for blender dataset (load train + val + test data), which is the default behavior of instant-ngp.
• 2.28: density_grid now stores density on the voxel center (with randomness), instead of on the grid. This should improve the rendering quality, such as the black strips in the lego scene.
• 2.23: better support for the blender dataset.
• 2.22: add GUI for NeRF training.
• 2.21: add GUI for NeRF visualizing.
  • With the GUI, I find the trained NeRF model is very noisy outside the seen region (unlike the original implementation)...
  • check mark_untrained_density_grid, but still, they looks much better even with the noises.
• 2.20: cuda raymarching is finally stable now!
• 2.15: add the official tinycudann as an alternative backend.
• 2.10: add cuda_ray, can train/infer faster, but performance is worse currently.
• 2.6: add support for RGBA image.
• 1.30: fixed atomicAdd() to use __half2 in HashGrid Encoder's backward, now the training speed with fp16 is as expected!
• 1.29:
  • finished an experimental binding of fully-fused MLP.
  • replace SHEncoder with a CUDA implementation.
• 1.26: add fp16 support for HashGrid Encoder (requires CUDA >= 10 and GPU ARCH >= 70 for now...).

Acknowledgement

• Credits to Thomas Müller for the amazing tiny-cuda-nn and instant-ngp:

  @misc{tiny-cuda-nn,
      Author = {Thomas M\"uller},
      Year = {2021},
      Note = {https://github.com/nvlabs/tiny-cuda-nn},
      Title = {Tiny {CUDA} Neural Network Framework}
  }
  
  @article{mueller2022instant,
      title = {Instant Neural Graphics Primitives with a Multiresolution Hash Encoding},
      author = {Thomas M\"uller and Alex Evans and Christoph Schied and Alexander Keller},
      journal = {arXiv:2201.05989},
      year = {2022},
      month = jan
  }
  

• The framework of NeRF is adapted from nerf_pl:

  @misc{queianchen_nerf,
      author = {Quei-An, Chen},
      title = {Nerf_pl: a pytorch-lightning implementation of NeRF},
      url = {https://github.com/kwea123/nerf_pl/},
      year = {2020},
  }
  

• The NeRF GUI is developed with DearPyGui.