Research map / 3D vision / NeRF and radiance fields
Geometry-Guided Progressive NeRF for Generalizable and Efficient Neural Human Rendering
Mingfei Chen, Jianfeng Zhang, Xiangyu Xu, Lijuan Liu, Yujun Cai, Jiashi Feng, Shuicheng Yan
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 |
|---|---|---|---|
| add geometry-guided attention to aggregate multi-view features (QG vs G) (ablation) | GP-NeRF · ZJU-MoCap uniform cameras | PSNR: 23.47 -> 23.68 | Improved |
| add pixel-aligned image features (P vs G) (ablation) | GP-NeRF · ZJU-MoCap uniform cameras | PSNR: 23.47 -> 26.09 | Improved |
| add geometry features QG to pixel-aligned features (QG+P vs P) (ablation) | GP-NeRF · ZJU-MoCap uniform cameras | PSNR: 26.09 -> 26.69 | Improved |
| add geometry features QG to pixel-aligned features (QG+P vs P) (ablation) | GP-NeRF · ZJU-MoCap uniform cameras | SSIM: 0.915 -> 0.924 | Improved |
| decouple density and appearance networks into progressive pipeline (35k steps) (ablation) | GP-NeRF · ZJU-MoCap uniform cameras | PSNR: 26.69 -> 26.65 | No clear effect |
| decouple density and appearance networks (5k steps) (ablation) | GP-NeRF · ZJU-MoCap uniform cameras | PSNR: 26.05 -> 26.13 | Improved |
| use P-only appearance network in progressive pipeline (5k steps) (ablation) | GP-NeRF · ZJU-MoCap uniform cameras | SSIM: 0.917 -> 0.920 | Improved |
| use QG+P instead of QG for density network (35k steps) (ablation) | GP-NeRF · ZJU-MoCap uniform cameras | PSNR: 26.40 -> 26.67 | Improved |
| use progressive rendering with fewer sampling points (GP-NeRF 1x vs NB 2x) | GP-NeRF · ZJU-MoCap | rendering time (ms): 611 -> 175 | Improved |
| use progressive rendering with fewer sampling points (GP-NeRF 2x vs NB 2x) | GP-NeRF · ZJU-MoCap | memory (GB): 21.80 -> 9.88 | Improved |
| add progressive rendering to GP-NeRF vs vanilla rendering (2x) | GP-NeRF · ZJU-MoCap | time (ms): 567 -> 243 | Improved |
| train on THUman-all instead of THUman-7 | GP-NeRF · ZJU-3 cross-dataset | PSNR: 24.74 -> 25.60 | Improved |
Closest papers by meaning
- GM-NeRF: Learning Generalizable Model-Based Neural Radiance Fields From Multi-View Images (CVPR 2023 · similarity 0.92)
- A Comprehensive Benchmark for Neural Human Radiance Fields (NeurIPS 2023 · similarity 0.92)
- H-NeRF: Neural Radiance Fields for Rendering and Temporal Reconstruction of Humans in Motion (NeurIPS 2021 · similarity 0.91)
- MPS-NeRF: Generalizable 3D Human Rendering From Multiview Images (TPAMI 2022 · similarity 0.90)
- Neural Capture of Animatable 3D Human from Monocular Video (ECCV 2022 · similarity 0.89)
- Real-time Free-view Human Rendering from Sparse-view RGB Videos using Double Unprojected Textures (CVPR 2025 · similarity 0.88)
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.
