Research map / 3D vision / Face and shape
Interpretable 3D Neural Object Volumes for Robust Conceptual Reasoning
Nhi Pham, Artur Jesslen, Bernt Schiele, Adam Kortylewski, Jonas Fischer
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 |
|---|---|---|---|
| replace CAVE with ground-truth pose by weak Orient-Anything pose supervision (ablation) | CAVE · OOD-CV | accuracy (%): 84.0 -> 80.4 | Hurt |
| replace CAVE with ground-truth pose by weak Orient-Anything pose supervision (ablation) | CAVE · Occluded P3D+ | accuracy (%): 81.5 -> 77.4 | Hurt |
| replace CAVE with ground-truth pose by weak Orient-Anything pose supervision (ablation) | CAVE · ImageNet3D | accuracy (%): 88.5 -> 84.7 | Hurt |
| replace dense NOVUM Gaussians (about 1130 per class) with 20 K-means concepts per class (ablation) | NOVUM · OOD-CV (weak pose) | accuracy (%): 79.7 -> 80.4 | Improved |
| replace dense NOVUM Gaussians with 20 concepts per class, ground-truth pose (ablation) | NOVUM · Pascal3D+ | accuracy (%): 99.5 -> 99.4 | No clear effect |
| replace dense NOVUM Gaussians with 20 concepts per class, weak pose (ablation) | NOVUM · ImageNet3D | accuracy (%): 85.7 -> 84.7 | Hurt |
| use CAVE concept-based NOVs instead of post-hoc concepts on NOVUM | NOVUM · OOD-CV | 3D consistency: 0.15 -> 0.24 | Improved |
| use CAVE instead of strongest prototype network TesNet | TesNet · ImageNet3D | accuracy (%): 77.9 -> 84.7 | Improved |
| use CAVE instead of LF-CBM | LF-CBM · OOD-CV | accuracy (%): 73.5 -> 80.4 | Improved |
| use CAVE instead of LF-CBM | LF-CBM · Pascal-Part | object coverage: 0.56 -> 0.80 | Improved |
| use NOV-aware LRP instead of vanilla LRP or GradCAM (ablation) | CAVE concept attribution · qualitative OOD (snow, occlusion) | attribution coherence | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- NOVUM — NOVUM: Neural Object Volumes for Robust Object Classification (ECCV 2024) · better on 2 of 4 shared results
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
- Interpretable3D: An Ad-Hoc Interpretable Classifier for 3D Point Clouds (AAAI 2024 · similarity 0.87)
- ImageNet3D: Towards General-Purpose Object-Level 3D Understanding (NeurIPS 2024 · similarity 0.86)
- Masked Point-Entity Contrast for Open-Vocabulary 3D Scene Understanding (CVPR 2025 · similarity 0.85)
- Cross-Modal and Uncertainty-Aware Agglomeration for Open-Vocabulary 3D Scene Understanding (CVPR 2025 · similarity 0.85)
- Towards 3D Objectness Learning in an Open World (NeurIPS 2025 · similarity 0.85)
- 3DCoMPaT200: Language Grounded Large-Scale 3D Vision Dataset for Compositional Recognition (NeurIPS 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.
