Research map / 3D vision / NeRF and radiance fields
Content-Aware Radiance Fields: Aligning Model Complexity with Scene Intricacy Through Learned Bitwidth Quantization
Weihang Liu, Xue Xian Zheng, Jingyi Yu, Xin Lou
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| learn bitwidth per layer and per scene with A-CAQ instead of fixed bitwidth for all (ablation) | Instant-NGP quantization · Synthetic-NeRF (MDL) | FQR: 14.00 -> 7.76 | Improved |
| add layer-wise quantization (Ours 1d vs fixed 1a) (ablation) | Instant-NGP quantization · Synthetic-NeRF (MDL) | PSNR: 31.56 -> 32.30 | Improved |
| add scene-wise bitwidth to layer-wise A-CAQ (ablation) | A-CAQ · Synthetic-NeRF (MDL) | FQR: 8.93 -> 7.76 | Improved |
| add bitwidth loss L_bit (adversarial bitwidth search) (ablation) | A-CAQ · Synthetic-NeRF (MGL) | FQR: 6.60 -> 5.33 | Improved |
| add QAT training (L_NeRF) on top of bitwidth search (ablation) | A-CAQ · Synthetic-NeRF (MGL) | FQR: 6.43 -> 5.33 | Improved |
| post-training with L_NeRF on fixed bitwidth (ablation) | PTQ baseline · Synthetic-NeRF (MGL) | PSNR: 22.29 -> 25.06 | Improved |
| A-CAQ vs LSQ+ under metric-guided (resource-constrained) learning | Instant-NGP · Synthetic-NeRF MGL 10^-3 | PSNR: 20.26 -> 25.96 | Improved |
| A-CAQ vs LSQ+ under high-fidelity MDL setting | Instant-NGP · Synthetic-NeRF MDL | PSNR: FQR 9.60 -> 7.76; PSNR 32.11 -> 32.00 | Mixed |
| A-CAQ quantization vs full-precision Instant-NGP | Instant-NGP · Synthetic-NeRF MDL | BitOps (T): 71.01 -> 6.55 | Improved |
Closest papers by meaning
- Taming Latent Diffusion Model for Neural Radiance Field Inpainting (ECCV 2024 · similarity 0.90)
- Strata-NeRF : Neural Radiance Fields for Stratified Scenes (ICCV 2023 · similarity 0.90)
- AligNeRF: High-Fidelity Neural Radiance Fields via Alignment-Aware Training (CVPR 2023 · similarity 0.89)
- Feature 3DGS: Supercharging 3D Gaussian Splatting to Enable Distilled Feature Fields (CVPR 2024 · similarity 0.89)
- Compact 3D Gaussian Representation for Radiance Field (CVPR 2024 · similarity 0.89)
- Mask-Based Modeling for Neural Radiance Fields (ICLR 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
