Research map / Video / Action recognition
Appearance-and-Relation Networks for Video Classification
Limin Wang, Wei Li, Wen Li, Luc Van Gool
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 |
|---|---|---|---|
| Replace 3D convolutions with SMART blocks | ResNet-18 on Kinetics · Kinetics val | average Top-1 and Top-5 accuracy: 75.7% -> 78.7% | Improved |
| Use a relation branch instead of 3D convolution (ablation) | ResNet-18 on Kinetics · Kinetics val | average Top-1 and Top-5 accuracy: 75.7% -> 77.2% | Improved |
| Use 2D convolutions instead of 3D convolutions (ablation) | ResNet-18 on Kinetics · Kinetics val | average Top-1 and Top-5 accuracy: 75.7% -> 71.9% | Hurt |
| Replace only the first 3D convolution with a SMART block (ablation) | ResNet-18 on Kinetics · Kinetics val | average Top-1 and Top-5 accuracy: 75.7% -> 77.4% | Improved |
| Stack multiple SMART blocks rather than use one (ablation) | ARTNet-ResNet18 on Kinetics · Kinetics val | average Top-1 and Top-5 accuracy: 77.4% -> 78.7% | Improved |
| Stack multiple relation branches (ablation) | Relation-ResNet18 on Kinetics · Kinetics val | average Top-1 and Top-5 accuracy: 77.2% -> 76.9% | Hurt |
| Add an optical-flow stream and fuse it with RGB (ablation) | C3D-ResNet18 on Kinetics · Kinetics val | average Top-1 and Top-5 accuracy: 75.7% -> 78.2% | Improved |
| Use RGB ARTNet instead of two-stream C3D | Kinetics validation · Kinetics val | average Top-1 and Top-5 accuracy: 78.2% -> 78.7% | Improved |
| Add an optical-flow stream and fuse it with RGB (ablation) | ARTNet-ResNet18 on Kinetics · Kinetics val | average Top-1 and Top-5 accuracy: 78.7% -> 80.4% | Improved |
| Add TSN long-term modeling (ablation) | ARTNet-ResNet18 with RGB on Kinetics · Kinetics val | average Top-1 and Top-5 accuracy: 78.7% -> 80.0% | Improved |
| Add TSN long-term modeling (ablation) | ARTNet-ResNet18 with Flow on Kinetics · Kinetics val | average Top-1 and Top-5 accuracy: 71.0% -> 71.9% | Improved |
| Add TSN long-term modeling (ablation) | ARTNet-ResNet18 with RGB and Flow on Kinetics · Kinetics val | average Top-1 and Top-5 accuracy: 80.4% -> 81.4% | Improved |
Closest papers by meaning
- Spatiotemporal Residual Networks for Video Action Recognition (NeurIPS 2016 · similarity 0.89)
- Deep Insights into Convolutional Networks for Video Recognition (IJCV 2019 · similarity 0.89)
- Collaborative Spatiotemporal Feature Learning for Video Action Recognition (CVPR 2019 · similarity 0.88)
- Rethinking Spatiotemporal Feature Learning: Speed-Accuracy Trade-offs in Video Classification (ECCV 2018 · similarity 0.88)
- Learning SpatioTemporal and Motion Features in a Unified 2D Network for Action Recognition (TPAMI 2022 · similarity 0.88)
- Video Modeling With Correlation Networks (CVPR 2020 · similarity 0.88)
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.
