# AttentionNAS: Spatiotemporal Attention Cell Search for Video Classification

arXiv: https://arxiv.org/abs/2007.12034 · HTML: https://mlsanity.com/paper/2007.12034/

## Key facts

- Cited by 23 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 23 Jul 2020.

## Built on by (most cited first, 23 of 23)

- [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)
- [VATT: Transformers for Multimodal Self-Supervised Learning from Raw Video, Audio and Text](https://mlsanity.com/paper/2104.11178/) (2021)
- [Keeping Your Eye on the Ball: Trajectory Attention in Video Transformers](https://mlsanity.com/paper/2106.05392/) (2021)
- [Adaptive Token Sampling For Efficient Vision Transformers](https://mlsanity.com/paper/2111.15667/) (2021)
- [Multiview Transformers for Video Recognition](https://mlsanity.com/paper/2201.04288/) (2022)
- [Space-time Mixing Attention for Video Transformer](https://mlsanity.com/paper/2106.05968/) (2021)
- [Video Transformers: A Survey](https://mlsanity.com/paper/2201.05991/) (2022)
- [PolyViT: Co-training Vision Transformers on Images, Videos and Audio](https://mlsanity.com/paper/2111.12993/) (2021)
- [MM-ViT: Multi-Modal Video Transformer for Compressed Video Action Recognition](https://mlsanity.com/paper/2108.09322/) (2021)
- [Shifted Chunk Transformer for Spatio-Temporal Representational Learning](https://mlsanity.com/paper/2108.11575/) (2021)
- [Video-FocalNets: Spatio-Temporal Focal Modulation for Video Action Recognition](https://mlsanity.com/paper/2307.06947/) (2023)
- [Gate-Shift-Fuse for Video Action Recognition](https://mlsanity.com/paper/2203.08897/) (2022)
- [MutualNet: Adaptive ConvNet via Mutual Learning from Different Model Configurations](https://mlsanity.com/paper/2105.07085/) (2021)
- [Generative Flows with Invertible Attentions](https://mlsanity.com/paper/2106.03959/) (2021)
- [Regional Attention with Architecture-Rebuilt 3D Network for RGB-D Gesture Recognition](https://mlsanity.com/paper/2102.05348/) (2021)
- [Video Violence Recognition and Localization Using a Semi-Supervised Hard Attention Model](https://mlsanity.com/paper/2202.02212/) (2022)
- [Two-Stream Transformer Architecture for Long Video Understanding](https://mlsanity.com/paper/2208.01753/) (2022)
- [VideoLightFormer: Lightweight Action Recognition using Transformers](https://mlsanity.com/paper/2107.00451/) (2021)
- [Weakly-supervised Representation Learning for Video Alignment and Analysis](https://mlsanity.com/paper/2302.04064/) (2023)
- [Searching for TrioNet: Combining Convolution with Local and Global Self-Attention](https://mlsanity.com/paper/2111.07547/) (2021)
- [StableMamba: Distillation-free Scaling of Large SSMs for Images and Videos](https://mlsanity.com/paper/2409.11867/) (2024)
- [Bringing together invertible UNets with invertible attention modules for memory-efficient diffusion models](https://mlsanity.com/paper/2504.10883/) (2025)

## Abstract (arXiv)

Convolutional operations have two limitations: (1) do not explicitly model where to focus as the same filter is applied to all the positions, and (2) are unsuitable for modeling long-range dependencies as they only operate on a small neighborhood. While both limitations can be alleviated by attention operations, many design choices remain to be determined to use attention, especially when applying attention to videos. Towards a principled way of applying attention to videos, we address the task of spatiotemporal attention cell search. We propose a novel search space for spatiotemporal attention cells, which allows the search algorithm to flexibly explore various design choices in the cell. The discovered attention cells can be seamlessly inserted into existing backbone networks, e.g., I3D or S3D, and improve video classification accuracy by more than 2% on both Kinetics-600 and MiT datasets. The discovered attention cells outperform non-local blocks on both datasets, and demonstrate strong generalization across different modalities, backbones, and datasets. Inserting our attention cells into I3D-R50 yields state-of-the-art performance on both datasets.