Research map / Video / Video segmentation and tracking
X3D: Expanding Architectures for Efficient Video Recognition
Check 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 |
|---|---|---|---|
| Progressively expand the tiny spatial model across selected axes | X3D · Kinetics-400 val, 10-Center | top-1 accuracy: 47.75% to 80.0% | Helped |
| Use progressive multi-axis expansion rather than direct 3D EfficientNet scaling | video recognition models · Kinetics-400 val | top-1 accuracy: +1.9 top-1; +2.2 top-1; +2.3 top-1 | Helped |
| Add channel-wise separable convolution while reducing bottleneck ratio to preserve compute (ablation) | X3D-S · Kinetics-400 val | top-1 accuracy: 68.9 -> 72.9 | Helped |
| Add channel-wise separable convolution while reducing bottleneck ratio to preserve compute (ablation) | X3D-XL · Kinetics-400 val | top-1 accuracy: 76.0 -> 78.4 | Helped |
| Replace channel-wise separable convolution with standard convolution at expanded bottleneck ratio (ablation) | X3D-S · Kinetics-400 val | top-1 accuracy: 72.9 -> 73.2 | Helped |
| Replace channel-wise separable convolution with standard convolution at expanded bottleneck ratio (ablation) | X3D-XL · Kinetics-400 val | top-1 accuracy: 78.4 -> 79.2 | Helped |
| Add swish nonlinearities (ablation) | X3D-S · Kinetics-400 val | top-1 accuracy: 72.0 -> 72.9 | Helped |
| Add swish nonlinearities (ablation) | X3D-XL · Kinetics-400 val | top-1 accuracy: 78.0 -> 78.4 | Helped |
| Add SE blocks (ablation) | X3D-S · Kinetics-400 val | top-1 accuracy: 71.3 -> 72.9 | Helped |
| Add SE blocks (ablation) | X3D-XL · Kinetics-400 val | top-1 accuracy: 77.1 -> 78.4 | Helped |
| Increase test-time clips from 1 to 3 (ablation) | video classifiers · Kinetics-400 val | top-1 accuracy | Helped |
| Increase test-time clips beyond 3 (ablation) | video classifiers · Kinetics-400 val | top-1 accuracy | No clear effect |
Closest papers by meaning
- MLP-3D: A MLP-Like 3D Architecture With Grouped Time Mixing (CVPR 2022 · similarity 0.87)
- 2D or not 2D? Adaptive 3D Convolution Selection for Efficient Video Recognition (CVPR 2021 · similarity 0.87)
- CAKES: Channel-wise Automatic KErnel Shrinking for Efficient 3D Networks (AAAI 2021 · similarity 0.86)
- Maximizing Spatio-Temporal Entropy of Deep 3D CNNs for Efficient Video Recognition (ICLR 2023 · similarity 0.86)
- X-3D: Explicit 3D Structure Modeling for Point Cloud Recognition (CVPR 2024 · similarity 0.84)
- OA-CNNs: Omni-Adaptive Sparse CNNs for 3D Semantic Segmentation (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.