Research map / 3D vision / Face and shape
Deep Learning on Object-Centric 3D Neural Fields
Pierluigi Zama Ramirez, Luca De Luigi, Daniele Sirocchi, Adriano Cardace, Riccardo Spezialetti, Francesco Ballerini, Samuele Salti, Luigi Di Stefano
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 |
|---|---|---|---|
| use ReLU instead of SIREN activation in input NFs (ablation) | nf2vec · ModelNet40 classification (UDF) | accuracy (%): 87.0 -> 88.1 | Improved |
| initialize all NFs with the same random vector (ablation) | nf2vec · NF weight alignment | convergence / linear mode connectivity | Improved |
| replace shared-network NFs (Functa) with individual NFs | NF reconstruction · Manifold40 mesh reconstruction | Chamfer distance (mm): 2.85 -> 0.26 | Improved |
| replace shared-network NFs (DeepSDF) with individual NFs | NF reconstruction · Manifold40 mesh reconstruction | F-score (%): 25.1 -> 69.7 | Improved |
| use the row-wise shared encoder instead of a flattened-weights MLP encoder | nf2vec encoder · 4-layer 512-unit NF | encoder parameters: ~800M -> ~3M | Improved |
| classify NeRF embeddings instead of rendered single view | ResNet50 · ShapeNetRender NeRF classification | accuracy (%): 86.88 -> 87.28 | No clear effect |
| classify NeRF embeddings instead of nine rendered views | ResNet50 multi-view · ShapeNetRender NeRF classification | accuracy (%): 93.28 -> 87.28 | Hurt |
| classify NeRF embeddings instead of rendering nine views | ResNet50 multi-view · NeRF classification inference | total time (ms): 97.56 -> 1.03 | Improved |
| interpolate nf2vec embeddings instead of NeRF weights | NeRF interpolation · ShapeNetRender interpolation | visual quality | Improved |
| use nf2vec embeddings instead of PointNet++ for retrieval | point cloud retrieval · ModelNet40/ShapeNet10/ScanNet10 | mAP: average gap of 1.8 mAP vs PointNet++ | Mixed |
| use nf2vec embeddings instead of DGCNN | point cloud classification · ModelNet40 | accuracy (%): 89.9 -> 87.0 | Hurt |
| use nf2vec instead of DWS on SDF classification | NF classification · Manifold40 (SDF) | accuracy (%): 69.9 -> 86.3 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DeepSDF — DeepSDF: Learning Continuous Signed Distance Functions for Shape Representation (CVPR 2019) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- How to Train Neural Field Representations: A Comprehensive Study and Benchmark (CVPR 2024 · similarity 0.89)
- Neural Unsigned Distance Fields for Implicit Function Learning (NeurIPS 2020 · similarity 0.88)
- N2F2: Hierarchical Scene Understanding with Nested Neural Feature Fields (ECCV 2024 · similarity 0.87)
- Neural Vector Fields for Implicit Surface Representation and Inference (IJCV 2024 · similarity 0.87)
- Neural Processing of Tri-Plane Hybrid Neural Fields (ICLR 2024 · similarity 0.86)
- ResFields: Residual Neural Fields for Spatiotemporal Signals (ICLR 2024 · similarity 0.85)
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.
