Research map / Architectures and efficiency / CNN design
RT3D: Achieving Real-Time Execution of 3D Convolutional Neural Networks on Mobile Devices
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 |
|---|---|---|---|
| Use reweighted regularization instead of fixed regularization (ablation) | C3D with KGS sparsity · UCF101 | Top-1 accuracy: 79.6% -> 80.5% | Helped |
| Use reweighted regularization instead of fixed regularization (ablation) | R(2+1)D with KGS sparsity · UCF101 | Top-1 accuracy: 91.7% -> 92.5% | Helped |
| Use KGS instead of Vanilla sparsity at the same pruning rate (ablation) | C3D with heuristic pruning · UCF101 | Top-1 accuracy: 78.8% -> 79.0% | Helped |
| Use KGS instead of Vanilla sparsity at the same pruning rate (ablation) | R(2+1)D with heuristic pruning · UCF101 | Top-1 accuracy: 89.4% -> 90.4% | Helped |
| Use KGS instead of Vanilla sparsity at the same pruning rate (ablation) | C3D with reweighted regularization, at equal pruning accuracy · UCF101 | FLOPs pruning rate: 2.4× -> 4.0× | Helped |
| Use KGS instead of Vanilla sparsity at the same pruning accuracy (ablation) | R(2+1)D with reweighted regularization · UCF101 | CPU latency: 523 -> 360 | Helped |
| Use KGS instead of Vanilla sparsity at the same pruning accuracy (ablation) | C3D with reweighted regularization · UCF101 | GPU latency: 236 -> 134 | Helped |
| Use KGS instead of Vanilla sparsity at the same pruning accuracy (ablation) | R(2+1)D with reweighted regularization · UCF101 | GPU latency: 225 -> 127 | Helped |
| Use reweighted regularization instead of heuristic pruning (ablation) | C3D with KGS sparsity · UCF101 | Top-1 accuracy: 79.0% -> 80.5% | Helped |
| Use reweighted regularization instead of heuristic pruning (ablation) | R(2+1)D with KGS sparsity · UCF101 | Top-1 accuracy: 90.4% -> 92.5% | Helped |
| Prune at the higher reported KGS FLOPs rate (ablation) | C3D with reweighted regularization · UCF101 | Top-1 accuracy: 80.5% -> 80.2% | Hurt |
| Prune at the higher reported KGS FLOPs rate (ablation) | R(2+1)D with reweighted regularization · UCF101 | Top-1 accuracy: 92.5% -> 92.0% | Hurt |
Closest papers by meaning
- PCONV: The Missing but Desirable Sparsity in DNN Weight Pruning for Real-Time Execution on Mobile Devices (AAAI 2020 · similarity 0.90)
- Faster CNNs with Direct Sparse Convolutions and Guided Pruning (ICLR 2017 · similarity 0.88)
- Accelerate CNNs from Three Dimensions: A Comprehensive Pruning Framework (ICML 2021 · similarity 0.87)
- DeepRebirth: Accelerating Deep Neural Network Execution on Mobile Devices (AAAI 2018 · similarity 0.87)
- MnasNet: Platform-Aware Neural Architecture Search for Mobile (CVPR 2019 · similarity 0.86)
- RepViT: Revisiting Mobile CNN From ViT Perspective (CVPR 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.