Research map / Architectures and efficiency / Pruning and efficient tuning
Spatial Re-Parameterization for N:M Sparsity
Yuxin Zhang, Mingbao Lin, Mingliang Xu, Yonghong Tian, Rongrong Ji
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 |
|---|---|---|---|
| add SpRe extra re-parameterized branch | SR-STE · ResNet-50 ImageNet-1K 1:16 | Top-1 accuracy: 71.5 -> 72.7 | Improved |
| add SpRe | LBC · ResNet-50 ImageNet-1K 1:16 | Top-1 accuracy: 71.8 -> 72.9 | Improved |
| add SpRe | ASP · ResNet-50 ImageNet-1K 1:16 | Top-1 accuracy: 71.5 -> 72.3 | Improved |
| add SpRe | ASP · MobileNet-V1 ImageNet-1K 1:16 | Top-1 accuracy: 50.4 -> 58.0 | Improved |
| add SpRe | SR-STE · MobileNet-V2 ImageNet-1K 1:16 | Top-1 accuracy: 24.2 -> 35.4 | Improved |
| add SpRe | SR-STE · ResNet-32 CIFAR-10 1:16 | Top-1 accuracy: 92.6 -> 93.9 | Improved |
| add SpRe at mild 2:4 sparsity | SR-STE · ResNet-32 CIFAR-10 2:4 | Top-1 accuracy: 94.9 -> 94.1 | Hurt |
| add SpRe | SR-STE · Faster R-CNN COCO 1:16 | mAP: 35.8 -> 36.3 | Improved |
| add SpRe | SR-STE · Mask R-CNN COCO 1:16 | Box mAP: 35.4 -> 36.1 | Improved |
| use same mask as main branch for extra branch (B^S = B) (ablation) | SR-STE w. SpRe · ResNet-50 ImageNet-1K | Top-1 accuracy | Hurt |
| allocate extra-branch weights inversely to unstructured spatial sparsity (ablation) | SR-STE w. SpRe · ResNet-50 ImageNet-1K | Top-1 accuracy | Hurt |
| add SpRe versus training baseline 60 more epochs (ablation) | ASP · ResNet-50 ImageNet-1K 1:16 | Top-1 accuracy: 71.8 -> 72.3 | Improved |
Closest papers by meaning
- Inference, Learning and Attention Mechanisms that Exploit and Preserve Sparsity in CNNs (IJCV 2020 · similarity 0.88)
- On Implicit Filter Level Sparsity in Convolutional Neural Networks (CVPR 2019 · similarity 0.87)
- Dynamic Spatial Sparsification for Efficient Vision Transformers and Convolutional Neural Networks (TPAMI 2023 · similarity 0.87)
- Sparsity in Deep Learning: Pruning and growth for efficient inference and training in neural networks (JMLR 2021 · similarity 0.87)
- Fast and Controllable Post-training Sparsity: Learning Optimal Sparsity Allocation with Global Constraint in Minutes (AAAI 2024 · similarity 0.86)
- Determining Layer-wise Sparsity for Large Language Models Through a Theoretical Perspective (ICML 2025 · 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.
