{"url":"/method/sta-lstm","slug":"sta-lstm","name":"STA-LSTM","full_name":"Spatio-Temporal Attention LSTM","full_name_withheld":false,"description_markdown":"In human action recognition, \r\neach type of action  generally only depends \r\non a few specific kinematic joints. Furthermore, over time, multiple actions may be performed.\r\nMotivated by these observations, Song et al. proposed \r\na joint spatial and temporal attention network based on LSTM, to adaptively find discriminative features and keyframes. \r\nIts main attention-related components are a spatial attention sub-network, to select important regions, and a temporal attention sub-network, to select key frames. The spatial attention sub-network can be written as:\r\n\\begin{align}\r\n    s_{t} &= U_{s}\\tanh(W_{xs}X_{t} + W_{hs}h_{t-1}^{s} + b_{si}) + b_{so}\r\n\\end{align}\r\n\\begin{align}\r\n    \\alpha_{t} &= \\text{Softmax}(s_{t})\r\n\\end{align}\r\n\\begin{align}\r\n    Y_{t} &= \\alpha_{t}  X_{t} \r\n\\end{align}\r\nwhere $X_{t}$ is the input feature at time $t$, $U_{s}$, $W_{hs}$, $b_{si}$, and $b_{so}$ are learnable parameters, and $h_{t-1}^{s}$ is the hidden state at step $t-1$. Note that use of the hidden state $h$ means  the attention process takes  temporal relationships into consideration.\r\n\r\nThe temporal attention sub-network is similar to the spatial branch and produces its attention map using:\r\n\\begin{align}\r\n    \\beta_{t} = \\delta(W_{xp}X_{t} + W_{hp}h_{t-1}^{p} + b_{p}). \r\n\\end{align}\r\nIt adopts a ReLU function instead of a normalization function for ease of optimization. It also uses a regularized objective function to improve  convergence.\r\n\r\nOverall, this paper presents a joint spatiotemporal attention method\r\nto focus on important joints and keyframes, \r\nwith excellent results on the action recognition task.","description_state":"present","introduced_year":null,"introduced_by":{"title":"An End-to-End Spatio-Temporal Attention Model for Human Action Recognition from Skeleton Data","paper":"/paper/an-end-to-end-spatio-temporal-attention-model","first_author":"Sijie Song","n_authors":5,"url_abs":null,"archive_paper_url":"https://paperswithcode.com/paper/an-end-to-end-spatio-temporal-attention-model"},"source":{"url":"http://arxiv.org/abs/1611.06067v1","title":"An End-to-End Spatio-Temporal Attention Model for Human Action Recognition from Skeleton Data","url_on_a_paper_host":true},"code_snippet_url":null,"code_snippet_url_on_a_code_host":false,"categories":[{"area":"General","area_id":"general","collection":"Attention Mechanisms","url":"/methods/category/attention-mechanisms","pwc_aliases":["attention-mechanisms-1"]}],"n_papers_tagged":2,"archive_num_papers":2,"papers_newest_first":[{"paper":null,"title":"Exploring Transformer-Augmented LSTM for Temporal and Spatial Feature Learning in Trajectory Prediction","date":"2024-12-18","arxiv_id":"2412.13419","n_code_links":0,"syntology":null},{"paper":"/paper/an-end-to-end-spatio-temporal-attention-model","title":"An End-to-End Spatio-Temporal Attention Model for Human Action Recognition from Skeleton Data","date":"2016-11-18","arxiv_id":"1611.06067","n_code_links":0,"syntology":null}],"papers_shown":2,"tasks":[{"task":"/task/action-recognition-in-videos","name":"Action Recognition","papers":1},{"task":"/task/autonomous-driving","name":"Autonomous Driving","papers":1},{"task":"/task/prediction","name":"Prediction","papers":1},{"task":"/task/skeleton-based-action-recognition","name":"Skeleton Based Action Recognition","papers":1},{"task":"/task/action-recognition","name":"Temporal Action Localization","papers":1},{"task":"/task/trajectory-prediction","name":"Trajectory Prediction","papers":1}],"tasks_shown":6,"n_tasks":6,"usage_by_year":[{"year":"2016","papers":1},{"year":"2024","papers":1}],"row_source":"methods_table","archive":{"source":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","archive_url":"https://paperswithcode.com/method/sta-lstm"},"syntology_read_at":"2026-09-24T18:15:14+00:00"}