Research map / 3D vision / NeRF and radiance fields
PeRFception: Perception using Radiance Fields
Yoonwoo Jeong, Seungjoo Shin, Junha Lee, Chris Choy, Anima Anandkumar, Minsu Cho, Jaesik Park
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 background augmentation with probability 0.5 (ablation) | ResNet18 · PeRFception-CO3D 2D classification | Acc@1 on PeRFception-CO3D: 81.43 -> 82.05 | Improved |
| add background augmentation with probability 0.5 (ablation) | ResNet152 · PeRFception-CO3D 2D classification | Acc@1 on PeRFception-CO3D: 84.35 -> 85.28 | Improved |
| raise background augmentation probability from 0.5 to 1.0 (ablation) | ResNet18 · PeRFception-CO3D 2D classification | Acc@1 on PeRFception-CO3D: 82.05 -> 81.31 | Hurt |
| raise background augmentation probability from 0.5 to 1.0 (ablation) | ResNet152 · PeRFception-CO3D 2D classification | Acc@1 on PeRFception-CO3D: 85.28 -> 84.06 | Hurt |
| train on PeRFception-CO3D (p=0.0) instead of CO3D images, test on PeRFception-CO3D | ResNet18 · CO3D 2D classification | Acc@1: 80.39 -> 81.43 | Improved |
| use ImageNet pretrained weights | ResNet50 · CO3D 2D classification | Acc@1 on CO3D: 86.83 -> 89.51 | Improved |
| use ImageNet pretrained weights with background augmentation p=0.5 | ResNet18 · PeRFception-CO3D 2D classification | Acc@1 on PeRFception-CO3D: 82.05 -> 84.97 | Improved |
| increase network depth from ResNet18 to ResNet152 | ResNet · CO3D 2D classification | Acc@1 on CO3D: 84.74 -> 87.75 | Improved |
| use density or spherical harmonics as 3D input features instead of ones | 3D ResNet · PeRFception-CO3D 3D classification | Acc@1/Acc@5 | Improved |
| use SH + D as input features for segmentation | Res16UNet · PeRFception-ScanNet segmentation | mIoU/mAcc | Improved |
| vary Plenoxels resolution among 128, 256 and 512 (ablation) | Plenoxels · PeRFception-ScanNet | Avg. PSNR | No clear effect |
| use original CO3D point-cloud vs voxel representation for storage | PeRFception-CO3D · CO3D | memory: 1.44TB -> 1.33TB | Improved |
Closest papers by meaning
- Neural Radiance Fields From Sparse RGB-D Images for High-Quality View Synthesis (TPAMI 2022 · similarity 0.89)
- Ref-NeRF: Structured View-Dependent Appearance for Neural Radiance Fields (CVPR 2022 · similarity 0.89)
- NeRFLight: Fast and Light Neural Radiance Fields Using a Shared Feature Grid (CVPR 2023 · similarity 0.89)
- ReNeRF: Relightable Neural Radiance Fields with Nearfield Lighting (ICCV 2023 · similarity 0.88)
- PERF: Panoramic Neural Radiance Field From a Single Panorama (TPAMI 2024 · similarity 0.88)
- NeRFrac: Neural Radiance Fields through Refractive Surface (ICCV 2023 · 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.
