Research map / Video / Action recognition
MVFNet: Multi-View Fusion Network for Efficient Video Recognition
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 |
|---|---|---|---|
| Add multi-view spatiotemporal modeling instead of α=0 C2D (ablation) | ResNet-50 MVFNet · Something-Something V1 | top-1 accuracy: 17.12 -> 50.48 | Helped |
| Add multi-view spatiotemporal modeling instead of α=0 C2D (ablation) | ResNet-50 MVFNet · Kinetics-400 | top-1 accuracy: 71.87 -> 74.21 | Helped |
| Use α=1/8, 1/4, or 1/2 rather than α=1/8 (ablation) | ResNet-50 MVFNet · Kinetics-400 | top-1 accuracy: 74.21; 74.18; 74.21 | No clear effect |
| Increase the multi-view channel proportion from α=1/8 to α=1/2 (ablation) | ResNet-50 MVFNet · Something-Something V1 | top-1 accuracy: 49.74 -> 50.48 | Helped |
| Add MVF blocks in res4-5 rather than only res5 (ablation) | ResNet-50 MVFNet · Something-Something V1 | top-1 accuracy: 46.02 -> 50.48 | Helped |
| Add MVF blocks in res2-5 instead of res4-5 (ablation) | ResNet-50 MVFNet · Kinetics-400 | top-1 accuracy: 74.21 -> 74.22 | No clear effect |
| Add height and width views alongside the temporal view (ablation) | ResNet-50 MVFNet · Something-Something V1 | top-1 accuracy: 49.13 -> 50.48 | Helped |
| Add height and width views alongside the temporal view (ablation) | ResNet-50 MVFNet · Kinetics-400 | top-1 accuracy: 73.72 -> 74.21 | Helped |
| Share convolution weights across the three views rather than learning independent weights (ablation) | ResNet-50 MVFNet · Something-Something V1 | top-1 accuracy: 50.48 -> 47.21 | Hurt |
| Share convolution weights across the three views rather than learning independent weights (ablation) | ResNet-50 MVFNet · Kinetics-400 | top-1 accuracy: 74.21 -> 73.81 | Hurt |
| Use MVF rather than C2D with the same MobileNet-V2 backbone (ablation) | MobileNet-V2 · Kinetics-400 | top-1 accuracy: 64.4 -> 67.5 | Helped |
| Use MVFNet rather than TSM | ResNet-50 · Kinetics-400 | top-1 accuracy: 74.1% vs. 76.0% | Helped |
Closest papers by meaning
- Convolutional Two-Stream Network Fusion for Video Action Recognition (CVPR 2016 · similarity 0.90)
- Spatiotemporal Pyramid Network for Video Action Recognition (CVPR 2017 · similarity 0.89)
- Deep Analysis of CNN-Based Spatio-Temporal Representations for Action Recognition (CVPR 2021 · similarity 0.88)
- Spatiotemporal Fusion in 3D CNNs: A Probabilistic View (CVPR 2020 · similarity 0.88)
- Spatiotemporal Residual Networks for Video Action Recognition (NeurIPS 2016 · similarity 0.88)
- DFDNet: Disentangling and Filtering Dynamics for Enhanced Video Prediction (AAAI 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.