# Relational Self-Attention: What's Missing in Attention for Video Understanding

arXiv: https://arxiv.org/abs/2111.01673 · HTML: https://mlsanity.com/paper/2111.01673/

## Key facts

- Cited by 15 papers in the map (data through 14 Oct 2026).
- Builds on 54 of its 66 references that are in the map.
- First posted on arXiv 2 Nov 2021.

## Builds on (most cited first, 30 of 54)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Deep Residual Learning for Image Recognition](https://mlsanity.com/paper/1512.03385/) (2015)
- [Very Deep Convolutional Networks for Large-Scale Image Recognition](https://mlsanity.com/paper/1409.1556/) (2014)
- [End-to-End Object Detection with Transformers](https://mlsanity.com/paper/2005.12872/) (2020)
- [EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks](https://mlsanity.com/paper/1905.11946/) (2019)
- [Rethinking the Inception Architecture for Computer Vision](https://mlsanity.com/paper/1512.00567/) (2015)
- [SGDR: Stochastic Gradient Descent with Warm Restarts](https://mlsanity.com/paper/1608.03983/) (2016)
- [MobileNetV2: Inverted Residuals and Linear Bottlenecks](https://mlsanity.com/paper/1801.04381/) (2018)
- [Going Deeper with Convolutions](https://mlsanity.com/paper/1409.4842/) (2014)
- [Delving Deep into Rectifiers: Surpassing Human-Level Performance on ImageNet Classification](https://mlsanity.com/paper/1502.01852/) (2015)
- [Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset](https://mlsanity.com/paper/1705.07750/) (2017)
- [Non-local Neural Networks](https://mlsanity.com/paper/1711.07971/) (2017)
- [SlowFast Networks for Video Recognition](https://mlsanity.com/paper/1812.03982/) (2018)
- [Is Space-Time Attention All You Need for Video Understanding?](https://mlsanity.com/paper/2102.05095/) (2021)
- [ViViT: A Video Vision Transformer](https://mlsanity.com/paper/2103.15691/) (2021)
- [A Closer Look at Spatiotemporal Convolutions for Action Recognition](https://mlsanity.com/paper/1711.11248/) (2017)
- [ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design](https://mlsanity.com/paper/1807.11164/) (2018)
- [The "something something" video database for learning and evaluating visual common sense](https://mlsanity.com/paper/1706.04261/) (2017)
- [Tokens-to-Token ViT: Training Vision Transformers from Scratch on ImageNet](https://mlsanity.com/paper/2101.11986/) (2021)
- [Two-Stream Convolutional Networks for Action Recognition in Videos](https://mlsanity.com/paper/1406.2199/) (2014)
- [Deep Networks with Stochastic Depth](https://mlsanity.com/paper/1603.09382/) (2016)
- [Temporal Segment Networks: Towards Good Practices for Deep Action Recognition](https://mlsanity.com/paper/1608.00859/) (2016)
- [TSM: Temporal Shift Module for Efficient Video Understanding](https://mlsanity.com/paper/1811.08383/) (2018)
- [Segmenter: Transformer for Semantic Segmentation](https://mlsanity.com/paper/2105.05633/) (2021)
- [X3D: Expanding Architectures for Efficient Video Recognition](https://mlsanity.com/paper/2004.04730/) (2020)
- [Stand-Alone Self-Attention in Vision Models](https://mlsanity.com/paper/1906.05909/) (2019)
- [Convolutional Two-Stream Network Fusion for Video Action Recognition](https://mlsanity.com/paper/1604.06573/) (2016)
- [Dynamic Convolution: Attention over Convolution Kernels](https://mlsanity.com/paper/1912.03458/) (2019)
- [Temporal Relational Reasoning in Videos](https://mlsanity.com/paper/1711.08496/) (2017)
- [Dynamic Filter Networks](https://mlsanity.com/paper/1605.09673/) (2016)

## Built on by (most cited first, 15 of 15)

- [Future Transformer for Long-term Action Anticipation](https://mlsanity.com/paper/2205.14022/) (2022)
- [MomentDiff: Generative Video Moment Retrieval from Random to Real](https://mlsanity.com/paper/2307.02869/) (2023)
- [Look Outside the Room: Synthesizing A Consistent Long-Term 3D Scene Video from A Single Image](https://mlsanity.com/paper/2203.09457/) (2022)
- [Video-FocalNets: Spatio-Temporal Focal Modulation for Video Action Recognition](https://mlsanity.com/paper/2307.06947/) (2023)
- [Learning Correlation Structures for Vision Transformers](https://mlsanity.com/paper/2404.03924/) (2024)
- [Tubelet-Contrastive Self-Supervision for Video-Efficient Generalization](https://mlsanity.com/paper/2303.11003/) (2023)
- [Uni-AdaFocus: Spatial-temporal Dynamic Computation for Video Recognition](https://mlsanity.com/paper/2412.11228/) (2024)
- [Learning Transferable Spatiotemporal Representations from Natural Script Knowledge](https://mlsanity.com/paper/2209.15280/) (2022)
- [LocoMotion: Learning Motion-Focused Video-Language Representations](https://mlsanity.com/paper/2410.12018/) (2024)
- [Sim-DETR: Unlock DETR for Temporal Sentence Grounding](https://mlsanity.com/paper/2509.23867/) (2025)
- [Relational Future Captioning Model for Explaining Likely Collisions in Daily Tasks](https://mlsanity.com/paper/2207.09083/) (2022)
- [Lightweight Structure-Aware Attention for Visual Understanding](https://mlsanity.com/paper/2211.16289/) (2022)
- [Causal Diffusion Models for Counterfactual Outcome Distributions in Longitudinal Data](https://mlsanity.com/paper/2604.12992/) (2026)
- [Exploring High-Order Self-Similarity for Video Understanding](https://mlsanity.com/paper/2604.20760/) (2026)
- [Tempered Self-Similarity Alignment for Physically Plausible Video Generation](https://mlsanity.com/paper/2605.24962/) (2026)

## Abstract (arXiv)

Convolution has been arguably the most important feature transform for modern neural networks, leading to the advance of deep learning. Recent emergence of Transformer networks, which replace convolution layers with self-attention blocks, has revealed the limitation of stationary convolution kernels and opened the door to the era of dynamic feature transforms. The existing dynamic transforms, including self-attention, however, are all limited for video understanding where correspondence relations in space and time, i.e., motion information, are crucial for effective representation. In this work, we introduce a relational feature transform, dubbed the relational self-attention (RSA), that leverages rich structures of spatio-temporal relations in videos by dynamically generating relational kernels and aggregating relational contexts. Our experiments and ablation studies show that the RSA network substantially outperforms convolution and self-attention counterparts, achieving the state of the art on the standard motion-centric benchmarks for video action recognition, such as Something-Something-V1 & V2, Diving48, and FineGym.