Research map / Architectures and efficiency / CNN design
Dynamic Group Convolution for Accelerating Convolutional Neural Networks
Zhuo Su, Linpu Fang, Wenxiong Kang, Dewen Hu, Matti Pietikäinen, Li Liu
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 dynamic per-sample channel selection (DGC) instead of fixed learned group conv (FLGC) | MobileNetV2 · CIFAR-10 | top-1 error (%): 5.89 -> 5.17 | Improved |
| use DGC instead of learned group convolution (LGC) | CondenseNet-50 · CIFAR-10 | top-1 error (%): 6.10 -> 5.83 | Improved |
| use DGC instead of LGC | CondenseNet-50 · CIFAR-100 | top-1 error (%): 26.58 -> 26.18 | Improved |
| use DGC instead of standard group conv (SGC) | CondenseNet · ImageNet | top-1 error (%): 29.0 -> 25.4 | Improved |
| use DGC instead of static filter pruning FPGM | ResNet-18 · ImageNet | top-1 error (%): 31.59 -> 31.22 | Improved |
| use DGC instead of dynamic channel pruning FBS | ResNet-18 · ImageNet | top-1 error (%): 31.83 -> 31.22 | Improved |
| use global threshold instead of head-wise threshold (DGC-G) | ResNet-18 · ImageNet | top-1 error (%): 31.22 -> 31.37 | Hurt |
| use global threshold instead of head-wise threshold (DGC-G) | CondenseNet · ImageNet | top-1 error (%): 25.4 -> 25.2 | Improved |
| use global threshold instead of head-wise threshold (DGC-G) | CondenseNet-86 · CIFAR-10 | top-1 error (%): 4.77 -> 4.42 | Improved |
| add angle enlargement loss with global threshold (ablation) | CondenseNet-DGC-G · ImageNet | top-1 error (%): 25.8 -> 25.2 | Improved |
| add angle enlargement loss with head-wise threshold (ablation) | DGC · ImageNet | top-1 error (%) | No clear effect |
| increase number of heads from 1 to 4, then 8 (ablation) | CondenseNet-DGC-50 · CIFAR-10 | top-1 error (%) | Mixed |
Closest papers by meaning
- Differentiable Learning-to-Group Channels via Groupable Convolutional Neural Networks (ICCV 2019 · similarity 0.93)
- Fully Learnable Group Convolution for Acceleration of Deep Neural Networks (CVPR 2019 · similarity 0.91)
- Dynamic Group Normalization: Spatio-Temporal Adaptation to Evolving Data Statistics (CVPR 2025 · similarity 0.87)
- Dynamic Clustering Convolutional Neural Network (AAAI 2025 · similarity 0.85)
- On the VC dimension of deep group convolutional neural networks (NeurIPS 2025 · similarity 0.85)
- KernelWarehouse: Rethinking the Design of Dynamic Convolution (ICML 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.
