Research map / 3D vision / Face and shape
Learning Feature Aggregation for Deep 3D Morphable Models
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Learn receptive fields and aggregation weights with attention instead of using fixed or directly parameterized mappings | Deep3DMM (spectral) · COMA | reconstruction error (mm): 0.931 -> 0.519 | Helped |
| Replace QEM mapping with attention-based feature aggregation | Deep3DMM (spectral), simple filters · COMA | reconstruction error: 0.939 -> 0.519 | Helped |
| Replace QEM mapping with attention-based feature aggregation | Deep3DMM (spectral), wider filters · COMA | reconstruction error: 0.682 -> 0.441 | Helped |
| Add feature aggregation to the encoder while keeping QEM in the decoder (ablation) | Deep3DMM (spectral) · COMA | reconstruction error (mm): 0.939 -> 0.886 | Helped |
| Add feature aggregation to the decoder while keeping QEM in the encoder (ablation) | Deep3DMM (spectral) · COMA | reconstruction error (mm): 0.939 -> 0.525 | Helped |
| Add feature aggregation to both encoder and decoder instead of only the decoder (ablation) | Deep3DMM (spectral) · COMA | reconstruction error (mm): 0.525 -> 0.519 | Helped |
| Fuse learned mapping matrices with precomputed mappings (ablation) | Deep3DMM (spectral) · COMA | reconstruction error: 0.693 -> 0.519 | Helped |
| Initialize keys and queries from template vertex positions rather than a normal distribution (ablation) | Deep3DMM (spectral) · COMA | reconstruction error (mm): 0.649 -> 0.519 | Helped |
| Initialize keys and queries from template vertex positions rather than a uniform distribution (ablation) | Deep3DMM (spectral) · COMA | reconstruction error (mm): 0.619 -> 0.519 | Helped |
| Use Chebyshev order K=6 instead of K=3 (ablation) | Deep3DMM (spectral) with learned feature aggregation · COMA | reconstruction error: 0.558 -> 0.519 | Helped |
| Apply top-k masking in the decoder (ablation) | Deep3DMM feature aggregation · COMA | reconstruction error | Helped |
| Apply top-k masking in the encoder (ablation) | Deep3DMM feature aggregation · COMA | reconstruction error | Hurt |
Closest papers by meaning
- Common3D: Self-Supervised Learning of 3D Morphable Models for Common Objects in Neural Feature Space (CVPR 2025 · similarity 0.89)
- Combining 3D Morphable Models: A Large Scale Face-And-Head Model (CVPR 2019 · similarity 0.88)
- Regressing Robust and Discriminative 3D Morphable Models With a Very Deep Neural Network (CVPR 2017 · similarity 0.88)
- Unsupervised Training for 3D Morphable Model Regression (CVPR 2018 · similarity 0.87)
- Locally Adaptive Neural 3D Morphable Models (CVPR 2024 · similarity 0.86)
- ImageNet3D: Towards General-Purpose Object-Level 3D Understanding (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.