Research map / Video / Action recognition
Interaction-aware Spatio-temporal Pyramid Attention Networks for Action Classification
Yang Du, Chunfeng Yuan, Bing Li, Lili Zhao, Yangxi Li, Weiming Hu
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 |
|---|---|---|---|
| use 3 pyramid scales instead of 1 scale (ablation) | Interaction-aware spatio-temporal pyramid attention · UCF101 split 1 (Inception-ResNet-V2) | accuracy RGB/Flow: 86.3% -> 87.3% (RGB), 84.0% -> 85.5% (Flow) | Improved |
| add a fourth scale (Conv2d_4a_3x3) to the 3-scale pyramid (ablation) | Interaction-aware spatio-temporal pyramid attention · UCF101 split 1 | accuracy RGB/Flow: 87.3% -> 86.5% (RGB), 85.5% -> 85.0% (Flow) | Hurt |
| aggregate the last conv layer of Block C instead of Block A (ablation) | Interaction-aware spatio-temporal pyramid attention · UCF101 split 1 | accuracy RGB/Flow: 85.8% -> 86.3% (RGB), 83.5% -> 84.0% (Flow) | Improved |
| aggregate a conv layer instead of the last fully-connected layer (ablation) | Interaction-aware spatio-temporal pyramid attention · UCF101 split 1 | accuracy RGB/Flow: 85.5% -> 86.3% (RGB), 83.4% -> 84.0% (Flow) | Improved |
| use element-wise multiplication instead of element-wise maximum for fusion (ablation) | Interaction-aware spatio-temporal pyramid attention · UCF101 split 1 | accuracy RGB: 85.7% -> 87.3% | Improved |
| use element-wise multiplication instead of element-wise sum for fusion (ablation) | Interaction-aware spatio-temporal pyramid attention · UCF101 split 1 | accuracy RGB: 86.4% -> 87.3% | Improved |
| add l_interactive and l_attn losses (ablation) | Interaction-aware spatio-temporal pyramid attention · UCF101 split 1 (3 scales) | accuracy RGB/Flow/Late fusion: 87.3% -> 88.2% (RGB), 85.5% -> 86.5% (Flow), 94.2% -> 95.1% (late fusion) | Improved |
| add only the l_interactive loss (ablation) | Interaction-aware spatio-temporal pyramid attention · UCF101 split 1 (3 scales) | accuracy RGB: 87.3% -> 87.8% | Improved |
| add only the l_attn loss (ablation) | Interaction-aware spatio-temporal pyramid attention · UCF101 split 1 (3 scales) | accuracy RGB: 87.3% -> 87.5% | Improved |
| plug the attention layer into VGGNet-16 | Two-Stream VGGNet-16 · UCF101 split 1 | accuracy late fusion: 90.7% -> 92.8% | Improved |
| plug the attention layer into BN-Inception | Two-Stream BN-Inception · UCF101 split 1 | accuracy late fusion: 92.0% -> 94.6% | Improved |
| plug the attention layer into Inception-ResNet-V2 | Two-Stream Inception-ResNet-V2 · UCF101 split 1 | accuracy late fusion: 92.6% -> 95.1% | Improved |
Closest papers by meaning
- AttentionNAS: Spatiotemporal Attention Cell Search for Video Classification (ECCV 2020 · similarity 0.88)
- Class Semantics-Based Attention for Action Detection (ICCV 2021 · similarity 0.87)
- Attentional Pooling for Action Recognition (NeurIPS 2017 · similarity 0.87)
- Explicitly Modeled Attention Maps for Image Classification (AAAI 2021 · similarity 0.87)
- An Efficient Spatio-Temporal Pyramid Transformer for Action Detection (ECCV 2022 · similarity 0.87)
- Spatiotemporal Pyramid Network for Video Action Recognition (CVPR 2017 · similarity 0.87)
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.
