Highlights

Rendering (during test-time) with OccGridEstimator is now 2x - 3x faster. (see #217).

What's Changed

• Remove unused code and fix issues when data is empty by @liruilong940607 in #211
• Accelerate Instant-NGP inference by @Linyou in #197
• Include a test mode rendering function for Instant NGP in the examples by @Linyou in #217
• fix: fix the dimension comments of t_sorted and t_indices by @lzhnb in #221
• Fix Workflows for Compiling by @liruilong940607 in #224

New Contributors

• @lzhnb made their first contribution in #221

Full Changelog: v0.5.2...v0.5.3

What's Changed

• fix issues in the example code by @FrozenSilent in #203
• Fixed grid traversal with near and far planes by @arterms in #204
• Per-ray minimum and maximum distances in sampling method by @arterms in #205
• Add pre-computed transmittance by @Linyou in #206

New Contributors

• @FrozenSilent made their first contribution in #203
• @arterms made their first contribution in #204
• @Linyou made their first contribution in #206

Full Changelog: v0.5.1...v0.5.2

Start supporting ray generation with CUDA acceleration.

• nerfacc.cameras.opencv_lens_undistortion
• nerfacc.cameras.opencv_lens_undistortion_fisheye

What's Changed

This is a major update with 90% of the underlying code rewritten. Aside from improved efficiency, we also have some methodology changes with more examples and benchmarks:

• Upgrade Occupancy Grid to support multiple levels.
• Support Proposal Network from Mip-NeRF 360.
• Update examples on unbounded scenes to use Multi-level Occupancy Grid or Proposal Network.
• Contraction for Occupancy Grid is no longer supported due to it's inefficiency for ray traversal.

Examples & Benchmarks:

Highlight

Multi-res Occupancy Grid is now support! Reduce the training time of 360 scenes from 20mins -> 5mins.

What's Changed

• Fix the link to train_mlp_nerf by @97littleleaf11 in #168
• Updates on examples by @97littleleaf11 in #174
• Support multi-res occ grid & prop net by @liruilong940607 in #176
• Fix train_mlp_nerf and save the model at the end of training by @97littleleaf11 in #177
• scatter_add_ -> index_add_ by @loicmagne in #179

New Contributors

• @97littleleaf11 made their first contribution in #168
• @loicmagne made their first contribution in #179

Full Changelog: v0.3.5...v0.4.0

Fix JIT build failure with pip install nerfacc.

Pre-built wheels on both Linux and Windows.

The easist way is to install from PyPI. In this way it will build the CUDA code on the first run (JIT).

pip install nerfacc

Or install from source. In this way it will build the CUDA code during installation.

pip install git+https://github.com/KAIR-BAIR/nerfacc.git

We also provide pre-built wheels covering major combinations of Pytorch + CUDA supported by official Pytorch.

# e.g., torch 1.13.0 + cu117
pip install nerfacc -f https://nerfacc-bucket.s3.us-west-2.amazonaws.com/whl/torch-1.13.0_cu117.html

* Pytorch does not support Windows pre-built wheels for those combinations thus we do not support as well.

What's Changed

** Minor bug fixes

What's Changed

TLDR: Minor fixes here and there.

• feat: add .pre-commit-config.yaml by @SauravMaheshkar in #118
• typos fixing by @Mirmix in #115
• Pass boolean tensor instead of the occlusion one. by @thomasw21 in #123
• Reduce precision conversion when packing by @thomasw21 in #124
• fixed a hang bug in ray_marching by @dozeri83 in #136
• rm build dir for first-time build by @liruilong940607 in #137

New Contributors

• @SauravMaheshkar made their first contribution in #118
• @Mirmix made their first contribution in #115
• @thomasw21 made their first contribution in #123
• @dozeri83 made their first contribution in #136

Full Changelog: v0.3.1...v0.3.2

What's Changed

• fix docs by @liruilong940607 in #105
• hotfix by @liruilong940607 in #109

Full Changelog: v0.3.0...v0.3.1