Research map / Architectures and efficiency / CNN design
Image as Set of Points
Read the paperCheck 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 |
|---|---|---|---|
| Add context cluster blocks to PointMLP | PointMLP · ScanObjectNN PB_T50_RS | mAcc(%): 83.9% to 84.4% (0.5%) | Helped |
| Add context cluster blocks to PointMLP | PointMLP · ScanObjectNN PB_T50_RS | OA(%): 85.4% to 86.2% (0.8%) | Helped |
| Use 25 centers per local region instead of 4 (ablation) | CoC-Small · MS COCO 2017 val, Mask R-CNN 1× | APbox: 35.9 -> 37.5 | Helped |
| Use 25 centers per local region instead of 4 (ablation) | CoC-Small · MS COCO 2017 val, Mask R-CNN 1× | APmask: 33.8 -> 35.4 | Helped |
| Increase centers per local region from 25 to 49 (ablation) | CoC-Small · MS COCO 2017 val, Mask R-CNN 1× | APbox: 37.5 -> 37.2 | Hurt |
| Use 25 centers per local region instead of 4 (ablation) | CoC-Small · ADE20K val, Semantic FPN | mIoU(%): 36.6 -> 36.4 | Hurt |
| Remove the context cluster operation (ablation) | Context-Cluster-Small · ImageNet-1K val | Top-1 accuracy: 77.5 -> 74.2 (3.3%) | Hurt |
| Add the multi-head design (ablation) | Context-Cluster-Small · ImageNet-1K val | Top-1 accuracy: 76.6 -> 77.5 (0.9%) | Helped |
| Remove region partitioning to allow global interactions (ablation) | CoC-Tiny · ImageNet-1K val | Top-1 accuracy: 71.8% -> 72.7% (0.9%) | Helped |
| Remove region partitioning to allow global interactions (ablation) | CoC-Tiny · ImageNet-1K | training memory: 23.39G -> 34.76G | Hurt |
| Update cluster centers twice instead of fixing them (ablation) | CoC-Tiny · ImageNet-1K | accuracy: 71.83% -> 71.85% (0.02%) | No clear effect |
| Change the region partition strategy to [49, 49, 1, 1] (ablation) | Context-Cluster-Ti · ImageNet-1K val | Top-1 accuracy: 71.8 -> 71.7 | No clear effect |
Closest papers by meaning
- Container: Context Aggregation Networks (NeurIPS 2021 · similarity 0.87)
- Dynamic Clustering Convolutional Neural Network (AAAI 2025 · similarity 0.86)
- Neural Clustering based Visual Representation Learning (CVPR 2024 · similarity 0.86)
- Revisiting the Integration of Convolution and Attention for Vision Backbone (NeurIPS 2024 · similarity 0.84)
- Image Clustering via the Principle of Rate Reduction in the Age of Pretrained Models (ICLR 2024 · similarity 0.84)
- MogaNet: Multi-order Gated Aggregation Network (ICLR 2024 · 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.