Research map / Point clouds / Point cloud classification
ACNe: Attentive Context Normalization for Robust Permutation-Equivariant Learning
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 |
|---|---|---|---|
| replace CNe with ACNe | CNe · MNIST point-cloud classification, 60% outliers | accuracy: 87.2 -> 93.7 | Helped |
| replace CNe with ACNe | CNe · Robust line fitting, 60% outliers | ℓ2 error: .00019 -> 1e-6 | Helped |
| add ACN to PointNet | PointNet · ModelNet40, 3D point-cloud classification | accuracy: 87.2 -> 87.7 | Helped |
| add ACNe in place of OANet | wide-baseline stereo pose estimation · Indoors | mAP at 10°/20°: .256/.392 -> .278/.428 | Helped |
| combine local and global attention (ablation) | ACNe · Saint Peter's Square stereo validation | mAP at 20°: .531 -> .597 | Helped |
| add supervision to local attention (ablation) | ACNe with local and global attention · Saint Peter's Square stereo validation | mAP at 20°: .597 -> .602 | Helped |
| use Group Normalization instead of Batch Normalization (ablation) | CNe · Saint Peter's Square stereo validation | mAP at 20°: .435 -> .414 | Hurt |
| apply attention directly to feature maps rather than to normalization (ablation) | ACNe · Saint Peter's Square stereo validation | mAP at 20°: .597 -> .408 | Hurt |
| use 3 ACN blocks per ARB instead of 2 (ablation) | ACNe · Saint Peter's Square stereo validation | mAP at 20°: .602 -> .621 | Helped |
| replace representation-to-weight learning with residual-to-weight learned IRLS (ablation) | ACNe · Robust line fitting, 70% outliers | ℓ2 error: .0008 -> .0024 | Hurt |
| add RANSAC post-processing | ACNe · Wide-baseline stereo pose estimation | mAP at 10°/20°: Outdoors .501/ .638 -> .478/.590; Indoors .278/.428 -> .209/.329 | Mixed |
| add RANSAC post-processing | ACNe · Essential matrix estimation, Outdoors | mAP at 20°: .706 -> .780 | Helped |
Closest papers by meaning
- Permuted AdaIN: Reducing the Bias Towards Global Statistics in Image Classification (CVPR 2021 · similarity 0.85)
- Towards More Robust Interpretation via Local Gradient Alignment (AAAI 2023 · similarity 0.85)
- Local Context Normalization: Revisiting Local Normalization (CVPR 2020 · similarity 0.85)
- Normalization-Equivariant Neural Networks with Application to Image Denoising (NeurIPS 2023 · similarity 0.84)
- Approximately Piecewise E(3) Equivariant Point Networks (ICLR 2024 · similarity 0.84)
- EquiTabPFN: A Target-Permutation Equivariant Prior Fitted Network (NeurIPS 2025 · similarity 0.84)
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.