Research map / 3D vision / NeRF and radiance fields
Weight Space Representation Learning on Diverse NeRF Architectures
Francesco Ballerini, Pierluigi Zama Ramirez, Luigi Di Stefano, Samuele Salti
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 contrastive (SigLIP) loss to rendering loss (ablation) | GMN encoder on ALL · cross-architecture classification (train MLP, test TRI) | accuracy: 25.3 -> 58.6 | Improved |
| add contrastive loss to rendering loss (ablation) | GMN encoder on ALL · cross-architecture classification (train TRI, test MLP) | accuracy: 11.4 -> 77.8 | Improved |
| add contrastive loss to rendering loss (ablation) | GMN encoder on ALL · ShapenetRender retrieval MLP/TRI Recall@1 | Recall@1 (%): 1.80 -> 30.62 | Improved |
| add contrastive loss to rendering loss (ablation) | GMN encoder on ALL · Objaverse retrieval MLP/TRI Recall@10 | Recall@10 (%): 9.66 -> 33.65 | Improved |
| add contrastive loss to rendering loss (ablation) | GMN encoder on ALL · classification, classifier trained on ALL, test MLP | accuracy: 93.6 -> 90.7 | Hurt |
| add contrastive loss to rendering loss (ablation) | GMN encoder on ALL · unseen TRI-16W classification | accuracy: 24.8 -> 72.1 | Improved |
| add contrastive loss to rendering loss (ablation) | GMN encoder on ALL · unseen TRI-8C classification (classifier on ALL) | accuracy: 69.4 -> 30.8 | Hurt |
| replace nf2vec with GMN encoder (rendering loss) on MLP NeRFs | GMN encoder · ShapenetRender classification (MLP) | accuracy: 92.1 -> 93.6 | Improved |
| replace tri-plane transformer of Cardace et al. with GMN encoder | GMN encoder · ShapenetRender classification (TRI) | accuracy: 93.1 -> 94.0 | Improved |
| use GMN encoder instead of nf2vec encoder in LLaNA | LLaNA · detailed captioning (MLP) | S-BERT: 77.4 -> 74.1 | Hurt |
| use GMN encoder instead of nf2vec encoder in LLaNA | LLaNA · Q&A (MLP) | S-BERT: 81.0 -> 81.0 | No clear effect |
| Cardace et al. vs GMN on unseen MLP-architecture changes | TRI-2L/TRI-32H · unseen classification | accuracy: 93.5 -> 92.6 (TRI-2L) | Hurt |
Closest papers by meaning
- Spatially-aware Weights Tokenization for NeRF-Language Models (NeurIPS 2025 · similarity 0.90)
- One Is All: Bridging the Gap between Neural Radiance Fields Architectures with Progressive Volume Distillation (AAAI 2023 · similarity 0.90)
- NeRFLight: Fast and Light Neural Radiance Fields Using a Shared Feature Grid (CVPR 2023 · similarity 0.90)
- D-NeRF: Neural Radiance Fields for Dynamic Scenes (CVPR 2021 · similarity 0.90)
- InsertNeRF: Instilling Generalizability into NeRF with HyperNet Modules (ICLR 2024 · similarity 0.89)
- LLaNA: Large Language and NeRF Assistant (NeurIPS 2024 · 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.
