Research map / Point clouds / Shape and surface completion
Embed Me if You Can: A Geometric Perceptron
Pavlo Melnyk, Michael Felsberg, Mårten Wadenbäck
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 MLGP point-wise embedding instead of MLP hidden layer | MLGP vs MLP · Tetris main dataset, noise a=0.0 | test accuracy (%): 71.7 -> 91.8 | Improved |
| use MLGP point-wise embedding instead of MLHP embedding | MLGP vs MLHP · Tetris main dataset, noise a=0.0 | test accuracy (%): 92.5 -> 91.8 | No clear effect |
| use MLGP instead of MLHP under noise a=0.2 | MLGP vs MLHP · Tetris main dataset, noise a=0.2 | test accuracy (%): 68.2 -> 69.7 | Improved |
| use MLGP instead of MLHP under noise a=0.2, top-10 by validation | MLGP vs MLHP · Tetris main dataset, noise a=0.2 | test accuracy (%): 72.2 -> 79.6 | Improved |
| use MLGP instead of MLHP on theta-split without noise | MLGP vs MLHP · Tetris theta-split, noise a=0.0 | test accuracy (%): 87.6 -> 87.9 | Improved |
| use MLGP instead of MLHP on theta-split noise a=0.1, top-10 | MLGP vs MLHP · Tetris theta-split, noise a=0.1 | test accuracy (%): 83.8 -> 87.5 | Improved |
| train/test on disjoint rotation ranges (theta-split) for MLP | MLP · Tetris theta-split vs main, noise a=0.0 | test accuracy (%): 71.7 -> 55.3 | Hurt |
| train/test on disjoint rotation ranges (theta-split) for MLGP | MLGP · Tetris theta-split vs main, noise a=0.0 | test accuracy (%): 91.8 -> 87.9 | Hurt |
| train/test on disjoint rotation ranges (theta-split) for MLHP | MLHP · Tetris theta-split vs main, noise a=0.0 | test accuracy (%): 92.5 -> 87.6 | Hurt |
| remove hidden-layer activation functions (identity) for MLGP and MLHP (ablation) | MLGP and MLHP · Tetris | test accuracy (%) | Improved |
| apply the same rigid transformation to geometric neuron weights as to the test data (ablation) | MLGP · Tetris isometry test | test accuracy (%): 86.09 -> 91.98 | Improved |
| transform only the data but not the weights (ablation) | MLGP · Tetris isometry test | test accuracy (%): 91.98 -> 86.09 | Hurt |
Closest papers by meaning
- TetraSphere: A Neural Descriptor for O(3)-Invariant Point Cloud Analysis (CVPR 2024 · similarity 0.84)
- Neural Processing of Tri-Plane Hybrid Neural Fields (ICLR 2024 · similarity 0.84)
- Improved MLP Point Cloud Processing with High-Dimensional Positional Encoding (AAAI 2024 · similarity 0.84)
- Hyperbolic Embeddings of Supervised Models (NeurIPS 2024 · similarity 0.84)
- Matrix Manifold Neural Networks++ (ICLR 2024 · similarity 0.83)
- Geometric Inductive Biases of Deep Networks: The Role of Data and Architecture (ICLR 2025 · similarity 0.83)
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.
