PosMLP-Video: Spatial and temporal relative position encoding for efficient video recognition
In recent years, vision Transformers and MLPs have demonstrated remarkable performance in image understanding tasks. However, their inherently dense computational operators, such as self-attention and token-mixing layers, pose significant challenges when applied to spatio-temporal video data. To add...
Saved in:
Main Authors: | , , , , , |
---|---|
Format: | text |
Language: | English |
Published: |
Institutional Knowledge at Singapore Management University
2023
|
Subjects: | |
Online Access: | https://ink.library.smu.edu.sg/sis_research/8256 https://ink.library.smu.edu.sg/context/sis_research/article/9259/viewcontent/PosMLP_preprint_pvoa_cc_by.pdf |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Singapore Management University |
Language: | English |
id |
sg-smu-ink.sis_research-9259 |
---|---|
record_format |
dspace |
spelling |
sg-smu-ink.sis_research-92592023-11-10T09:00:15Z PosMLP-Video: Spatial and temporal relative position encoding for efficient video recognition HAO, Yanbin ZHOU, Diansong WANG, Zhicai NGO, Chong-wah HE, Xiangnan WANG, Meng In recent years, vision Transformers and MLPs have demonstrated remarkable performance in image understanding tasks. However, their inherently dense computational operators, such as self-attention and token-mixing layers, pose significant challenges when applied to spatio-temporal video data. To address this gap, we propose PosMLP-Video, a lightweight yet powerful MLP-like backbone for video recognition. Instead of dense operators, we use efficient relative positional encoding (RPE) to build pairwise token relations, leveraging small-sized parameterized relative position biases to obtain each relation score. Specifically, to enable spatio-temporal modeling, we extend the image PosMLP’s positional gating unit to temporal, spatial, and spatio-temporal variants, namely PoTGU, PoSGU, and PoSTGU, respectively. These gating units can be feasibly combined into three types of spatio-temporal factorized positional MLP blocks, which not only decrease model complexity but also maintain good performance. Additionally, we improve the locality of modeling using window partitioning and enrich relative positional relationships using channel grouping. Experimental results demonstrate that PosMLP-Video achieves competitive speed-accuracy trade-offs compared to the previous state-of-the-art models. In particular, PosMLP-Video pre-trained on ImageNet1K achieves 59.0%/70.3% top-1 accuracy on Something-Something V1/V2 and 82.1% top-1 accuracy on Kinetics-400 while requiring much fewer parameters and FLOPs than other models. The code will be made publicly available. 2023-10-01T07:00:00Z text application/pdf https://ink.library.smu.edu.sg/sis_research/8256 info:doi/10.21203/rs.3.rs-3485088/v1 https://ink.library.smu.edu.sg/context/sis_research/article/9259/viewcontent/PosMLP_preprint_pvoa_cc_by.pdf http://creativecommons.org/licenses/by/3.0/ Research Collection School Of Computing and Information Systems eng Institutional Knowledge at Singapore Management University Positional encoding spatio-temporal modeling multi-layer perceptron video recognition Artificial Intelligence and Robotics Graphics and Human Computer Interfaces |
institution |
Singapore Management University |
building |
SMU Libraries |
continent |
Asia |
country |
Singapore Singapore |
content_provider |
SMU Libraries |
collection |
InK@SMU |
language |
English |
topic |
Positional encoding spatio-temporal modeling multi-layer perceptron video recognition Artificial Intelligence and Robotics Graphics and Human Computer Interfaces |
spellingShingle |
Positional encoding spatio-temporal modeling multi-layer perceptron video recognition Artificial Intelligence and Robotics Graphics and Human Computer Interfaces HAO, Yanbin ZHOU, Diansong WANG, Zhicai NGO, Chong-wah HE, Xiangnan WANG, Meng PosMLP-Video: Spatial and temporal relative position encoding for efficient video recognition |
description |
In recent years, vision Transformers and MLPs have demonstrated remarkable performance in image understanding tasks. However, their inherently dense computational operators, such as self-attention and token-mixing layers, pose significant challenges when applied to spatio-temporal video data. To address this gap, we propose PosMLP-Video, a lightweight yet powerful MLP-like backbone for video recognition. Instead of dense operators, we use efficient relative positional encoding (RPE) to build pairwise token relations, leveraging small-sized parameterized relative position biases to obtain each relation score. Specifically, to enable spatio-temporal modeling, we extend the image PosMLP’s positional gating unit to temporal, spatial, and spatio-temporal variants, namely PoTGU, PoSGU, and PoSTGU, respectively. These gating units can be feasibly combined into three types of spatio-temporal factorized positional MLP blocks, which not only decrease model complexity but also maintain good performance. Additionally, we improve the locality of modeling using window partitioning and enrich relative positional relationships using channel grouping. Experimental results demonstrate that PosMLP-Video achieves competitive speed-accuracy trade-offs compared to the previous state-of-the-art models. In particular, PosMLP-Video pre-trained on ImageNet1K achieves 59.0%/70.3% top-1 accuracy on Something-Something V1/V2 and 82.1% top-1 accuracy on Kinetics-400 while requiring much fewer parameters and FLOPs than other models. The code will be made publicly available. |
format |
text |
author |
HAO, Yanbin ZHOU, Diansong WANG, Zhicai NGO, Chong-wah HE, Xiangnan WANG, Meng |
author_facet |
HAO, Yanbin ZHOU, Diansong WANG, Zhicai NGO, Chong-wah HE, Xiangnan WANG, Meng |
author_sort |
HAO, Yanbin |
title |
PosMLP-Video: Spatial and temporal relative position encoding for efficient video recognition |
title_short |
PosMLP-Video: Spatial and temporal relative position encoding for efficient video recognition |
title_full |
PosMLP-Video: Spatial and temporal relative position encoding for efficient video recognition |
title_fullStr |
PosMLP-Video: Spatial and temporal relative position encoding for efficient video recognition |
title_full_unstemmed |
PosMLP-Video: Spatial and temporal relative position encoding for efficient video recognition |
title_sort |
posmlp-video: spatial and temporal relative position encoding for efficient video recognition |
publisher |
Institutional Knowledge at Singapore Management University |
publishDate |
2023 |
url |
https://ink.library.smu.edu.sg/sis_research/8256 https://ink.library.smu.edu.sg/context/sis_research/article/9259/viewcontent/PosMLP_preprint_pvoa_cc_by.pdf |
_version_ |
1783955658669817856 |