Research map / Video / Action recognition
A Semantic and Motion-Aware Spatiotemporal Transformer Network for Action Detection
Matthew Korban, Peter A. Youngs, Scott T. Acton
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 |
|---|---|---|---|
| add motion enhancement to motion-aware 2D positional encoding (ablation) | SMAST · AVA 2.1 | mAP@0.5 (%): 31.0 -> 33.1 | Improved |
| use motion-aware 2D PE without motion enhancement vs standard 1D PE (ablation) | SMAST · AVA 2.1 | mAP@0.5 (%): 30.9 -> 31.0 | No clear effect |
| add inter-feature attention A_GM to A_GG+A_MM (ablation) | SMAST · AVA 2.1 | mAP@0.5 (%): 28.0 -> 31.8 | Improved |
| add inter-feature attention A_MG to A_GG+A_MM (ablation) | SMAST · AVA 2.1 | mAP@0.5 (%): 28.0 -> 30.9 | Improved |
| add both A_GM and A_MG (full multi-feature attention) (ablation) | SMAST · AVA 2.1 | mAP@0.5 (%): 28.0 -> 33.1 | Improved |
| use inter-feature attention alone (ablation) | SMAST · AVA 2.1 | mAP@0.5 (%): 26.5 -> 24.4 | Hurt |
| use selective attention (Q persons, K persons+objects) instead of self-attention (ablation) | SMAST · AVA 2.1 | mAP@0.5 (%): 31.8 -> 33.1 | Improved |
| restrict self-attention to objects only vs persons+objects (ablation) | SMAST · AVA 2.1 | mAP@0.5 (%): 31.8 -> 24.3 | Hurt |
| use sequence-based temporal attention instead of efficient temporal attention (ablation) | SMAST · AVA 2.1 | mAP@0.5 (%): 30.9 -> 32.5 | Improved |
| add sequence-based temporal positional encoding PTPE (ablation) | SMAST · AVA 2.1 | mAP@0.5 (%): 32.5 -> 33.1 | Improved |
| efficient temporal attention instead of standard (ablation) | SMAST · AVA 2.1 | mAP@0.5 (%): 30.4 -> 30.9 | No clear effect |
| plug multi-feature attention into AIA | AIA · AVA 2.1 | mAP@0.5 (%): 31.2 -> 32.4 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- TubeR — TubeR: Tubelet Transformer for Video Action Detection (CVPR 2022) · better on 1 of 1 shared result
- VideoMAE — VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training (NeurIPS 2022) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Shrinking Temporal Attention in Transformers for Video Action Recognition (AAAI 2022 · similarity 0.89)
- Scene-Aware Spatiotemporal Generalization: Towards Robust Temporal Action Detection Across Domains (AAAI 2026 · similarity 0.89)
- Building a Multi-modal Spatiotemporal Expert for Zero-shot Action Recognition with CLIP (AAAI 2025 · similarity 0.88)
- An Efficient Spatio-Temporal Pyramid Transformer for Action Detection (ECCV 2022 · similarity 0.88)
- Video Action Recognition with Attentive Semantic Units (ICCV 2023 · similarity 0.87)
- Kronecker Mask and Interpretive Prompts are Language-Action Video Learners (ICLR 2025 · 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.
