A-CAQ for radiance field

⭐: Extended Fully Content Aware Framework is submitted to TVCG, codes will be released soon.

⭐: Paper is accepted by ECCV 2024 ArXiv Link.

---

This is the reference code for paper "Content-Aware Radiance Fields: Aligning Model Complexity with Scene Intricacy Through Learned Bitwidth Quantization"

Supported public datasets should be initially downloaded from the internet

• Synthetic-NeRF

• Mip-NeRF360 (COLMAP required)

• RTMV

  and put in folder ./data

Install

pip install -r requirements.txt

# install all extension modules
bash scripts/install_ext.sh
cd raymarching
python setup.py build_ext --inplace # build ext only, do not install (only can be used in the parent directory)
pip install . # install to python path (you still need the raymarching/ folder, since this only install the built extension.)

Tested environments

Ubuntu 20.04 with torch 2.0.0 & CUDA 11.8 on a RTX 4090

Usage

Examples see train.sh, train_penalty.sh, train_PTQ_LSQ.sh

To find the parameter definition, see ./config/GetConfig.py