Research map / Architectures and efficiency / CNN design
UniConvNet: Expanding Effective Receptive Field while Maintaining Asymptotically Gaussian Distribution for ConvNets of Any Scale
Yuhao Wang, Wei Xi
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 progressive kernels (7,9,11) instead of (5,7,9) (ablation) | UniConvNet-A · ImageNet-1K val, UniConvNet-A kernel ablation | top-1 accuracy: 76.6 -> 77.0 | Improved |
| use (7,9,11) instead of (9,11,13) (ablation) | UniConvNet-A · ImageNet-1K val, UniConvNet-A kernel ablation | top-1 accuracy: 76.9 -> 77.0 | Improved |
| use (9,11,13) instead of (7,9,11) (ablation) | UniConvNet-P0 · ImageNet-1K val | top-1 accuracy: 79.1 -> 79.3 | Mixed |
| use three progressive kernels (7,9,11) instead of (7,11,31) (ablation) | UniConvNet-A · ImageNet-1K val, UniConvNet-A kernel ablation | top-1 accuracy: 76.0 -> 77.0 | Improved |
| use (7,9,11) instead of (7,9,29) (ablation) | UniConvNet-A · ImageNet-1K val, UniConvNet-A kernel ablation | top-1 accuracy: 75.9 -> 77.0 | Improved |
| use (7,9,11) instead of (27,29,31) (ablation) | UniConvNet-N2 · ImageNet-1K val | top-1 accuracy: 81.8 -> 82.7 | Improved |
| add Three-layer RFA large-kernel to DW3x3 model (ablation) | UniConvNet-P0 · ImageNet-1K val | top-1 accuracy: 77.0 -> 78.9 | Improved |
| add modified DCNV3 to Three-layer RFA (ablation) | UniConvNet-P0 · ImageNet-1K val | top-1 accuracy: 78.4 -> 79.1 | Improved |
| use modified DCNV3 instead of DW 3x3 (ablation) | UniConvNet-P0 · ImageNet-1K val | top-1 accuracy: 78.9 -> 79.1 | Improved |
| use Three-layer RFA instead of DW 7x7 (ablation) | UniConvNet-P0 · ImageNet-1K val | top-1 accuracy: 77.0 -> 79.1 | Improved |
| use Three-layer RFA instead of DW 7x7 (ablation) | UniConvNet-N2 · ImageNet-1K val | top-1 accuracy: 81.0 -> 82.7 | Improved |
| add Three-layer RFA to 30M DW 3x3 model (ablation) | UniConvNet-T · ImageNet-1K val | top-1 accuracy: 82.1 -> 84.2 | Improved |
Closest papers by meaning
- UniRepLKNet: A Universal Perception Large-Kernel ConvNet for Audio Video Point Cloud Time-Series and Image Recognition (CVPR 2024 · similarity 0.90)
- Scaling Up Your Kernels: Large Kernel Design in ConvNets Toward Universal Representations (TPAMI 2025 · similarity 0.90)
- Receptive Fields As Experts in Convolutional Neural Architectures (ICML 2024 · similarity 0.88)
- AsCAN: Asymmetric Convolution-Attention Networks for Efficient Recognition and Generation (NeurIPS 2024 · similarity 0.86)
- PeLK: Parameter-efficient Large Kernel ConvNets with Peripheral Convolution (CVPR 2024 · similarity 0.86)
- KernelWarehouse: Rethinking the Design of Dynamic Convolution (ICML 2024 · similarity 0.86)
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.
