Token Pruning using a Lightweight Background Aware Vision Transformer

High runtime memory and high latency puts significant constraint on Vision Transformer training and inference, especially on edge devices. Token pruning reduces the number of input tokens to the ViT based on importance criteria of each token. We present a Background Aware Vision Transformer (BAViT)...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:arXiv.org 2024-10
Hauptverfasser: Sah, Sudhakar, Kumar, Ravish, Rohmetra, Honnesh, Saboori, Ehsan
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title arXiv.org
container_volume
creator Sah, Sudhakar
Kumar, Ravish
Rohmetra, Honnesh
Saboori, Ehsan
description High runtime memory and high latency puts significant constraint on Vision Transformer training and inference, especially on edge devices. Token pruning reduces the number of input tokens to the ViT based on importance criteria of each token. We present a Background Aware Vision Transformer (BAViT) model, a pre-processing block to object detection models like DETR/YOLOS aimed to reduce runtime memory and increase throughput by using a novel approach to identify background tokens in the image. The background tokens can be pruned completely or partially before feeding to a ViT based object detector. We use the semantic information provided by segmentation map and/or bounding box annotation to train a few layers of ViT to classify tokens to either foreground or background. Using 2 layers and 10 layers of BAViT, background and foreground tokens can be separated with 75% and 88% accuracy on VOC dataset and 71% and 80% accuracy on COCO dataset respectively. We show a 2 layer BAViT-small model as pre-processor to YOLOS can increase the throughput by 30% - 40% with a mAP drop of 3% without any sparse fine-tuning and 2% with sparse fine-tuning. Our approach is specifically targeted for Edge AI use cases.
format Article
fullrecord <record><control><sourceid>proquest</sourceid><recordid>TN_cdi_proquest_journals_3116744293</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>3116744293</sourcerecordid><originalsourceid>FETCH-proquest_journals_31167442933</originalsourceid><addsrcrecordid>eNqNissKwjAQAIMgWLT_sOC50CZ96FGl4sGDh-K1BE1rWt3obkN_XwU_wMvMYWYiAqlUEq1SKWciZO7iOJZ5IbNMBaKsXG8QTuTRYguev9RwtO1tGM2XsNWXviXn8QqbUZOBs2XrECrSyI2jh6GFmDb6zib8eS6W-7LaHaInuZc3PNSd84SfVKskyYs0lWul_rvef847AA</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>3116744293</pqid></control><display><type>article</type><title>Token Pruning using a Lightweight Background Aware Vision Transformer</title><source>Free E- Journals</source><creator>Sah, Sudhakar ; Kumar, Ravish ; Rohmetra, Honnesh ; Saboori, Ehsan</creator><creatorcontrib>Sah, Sudhakar ; Kumar, Ravish ; Rohmetra, Honnesh ; Saboori, Ehsan</creatorcontrib><description>High runtime memory and high latency puts significant constraint on Vision Transformer training and inference, especially on edge devices. Token pruning reduces the number of input tokens to the ViT based on importance criteria of each token. We present a Background Aware Vision Transformer (BAViT) model, a pre-processing block to object detection models like DETR/YOLOS aimed to reduce runtime memory and increase throughput by using a novel approach to identify background tokens in the image. The background tokens can be pruned completely or partially before feeding to a ViT based object detector. We use the semantic information provided by segmentation map and/or bounding box annotation to train a few layers of ViT to classify tokens to either foreground or background. Using 2 layers and 10 layers of BAViT, background and foreground tokens can be separated with 75% and 88% accuracy on VOC dataset and 71% and 80% accuracy on COCO dataset respectively. We show a 2 layer BAViT-small model as pre-processor to YOLOS can increase the throughput by 30% - 40% with a mAP drop of 3% without any sparse fine-tuning and 2% with sparse fine-tuning. Our approach is specifically targeted for Edge AI use cases.</description><identifier>EISSN: 2331-8422</identifier><language>eng</language><publisher>Ithaca: Cornell University Library, arXiv.org</publisher><subject>Accuracy ; Annotations ; Datasets ; Image segmentation ; Memory devices ; Microprocessors ; Object recognition ; Pruning</subject><ispartof>arXiv.org, 2024-10</ispartof><rights>2024. This work is published under http://creativecommons.org/licenses/by/4.0/ (the “License”). Notwithstanding the ProQuest Terms and Conditions, you may use this content in accordance with the terms of the License.</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>776,780</link.rule.ids></links><search><creatorcontrib>Sah, Sudhakar</creatorcontrib><creatorcontrib>Kumar, Ravish</creatorcontrib><creatorcontrib>Rohmetra, Honnesh</creatorcontrib><creatorcontrib>Saboori, Ehsan</creatorcontrib><title>Token Pruning using a Lightweight Background Aware Vision Transformer</title><title>arXiv.org</title><description>High runtime memory and high latency puts significant constraint on Vision Transformer training and inference, especially on edge devices. Token pruning reduces the number of input tokens to the ViT based on importance criteria of each token. We present a Background Aware Vision Transformer (BAViT) model, a pre-processing block to object detection models like DETR/YOLOS aimed to reduce runtime memory and increase throughput by using a novel approach to identify background tokens in the image. The background tokens can be pruned completely or partially before feeding to a ViT based object detector. We use the semantic information provided by segmentation map and/or bounding box annotation to train a few layers of ViT to classify tokens to either foreground or background. Using 2 layers and 10 layers of BAViT, background and foreground tokens can be separated with 75% and 88% accuracy on VOC dataset and 71% and 80% accuracy on COCO dataset respectively. We show a 2 layer BAViT-small model as pre-processor to YOLOS can increase the throughput by 30% - 40% with a mAP drop of 3% without any sparse fine-tuning and 2% with sparse fine-tuning. Our approach is specifically targeted for Edge AI use cases.</description><subject>Accuracy</subject><subject>Annotations</subject><subject>Datasets</subject><subject>Image segmentation</subject><subject>Memory devices</subject><subject>Microprocessors</subject><subject>Object recognition</subject><subject>Pruning</subject><issn>2331-8422</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2024</creationdate><recordtype>article</recordtype><sourceid>ABUWG</sourceid><sourceid>AFKRA</sourceid><sourceid>AZQEC</sourceid><sourceid>BENPR</sourceid><sourceid>CCPQU</sourceid><sourceid>DWQXO</sourceid><recordid>eNqNissKwjAQAIMgWLT_sOC50CZ96FGl4sGDh-K1BE1rWt3obkN_XwU_wMvMYWYiAqlUEq1SKWciZO7iOJZ5IbNMBaKsXG8QTuTRYguev9RwtO1tGM2XsNWXviXn8QqbUZOBs2XrECrSyI2jh6GFmDb6zib8eS6W-7LaHaInuZc3PNSd84SfVKskyYs0lWul_rvef847AA</recordid><startdate>20241012</startdate><enddate>20241012</enddate><creator>Sah, Sudhakar</creator><creator>Kumar, Ravish</creator><creator>Rohmetra, Honnesh</creator><creator>Saboori, Ehsan</creator><general>Cornell University Library, arXiv.org</general><scope>8FE</scope><scope>8FG</scope><scope>ABJCF</scope><scope>ABUWG</scope><scope>AFKRA</scope><scope>AZQEC</scope><scope>BENPR</scope><scope>BGLVJ</scope><scope>CCPQU</scope><scope>DWQXO</scope><scope>HCIFZ</scope><scope>L6V</scope><scope>M7S</scope><scope>PIMPY</scope><scope>PQEST</scope><scope>PQQKQ</scope><scope>PQUKI</scope><scope>PRINS</scope><scope>PTHSS</scope></search><sort><creationdate>20241012</creationdate><title>Token Pruning using a Lightweight Background Aware Vision Transformer</title><author>Sah, Sudhakar ; Kumar, Ravish ; Rohmetra, Honnesh ; Saboori, Ehsan</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-proquest_journals_31167442933</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2024</creationdate><topic>Accuracy</topic><topic>Annotations</topic><topic>Datasets</topic><topic>Image segmentation</topic><topic>Memory devices</topic><topic>Microprocessors</topic><topic>Object recognition</topic><topic>Pruning</topic><toplevel>online_resources</toplevel><creatorcontrib>Sah, Sudhakar</creatorcontrib><creatorcontrib>Kumar, Ravish</creatorcontrib><creatorcontrib>Rohmetra, Honnesh</creatorcontrib><creatorcontrib>Saboori, Ehsan</creatorcontrib><collection>ProQuest SciTech Collection</collection><collection>ProQuest Technology Collection</collection><collection>Materials Science &amp; Engineering Collection</collection><collection>ProQuest Central (Alumni Edition)</collection><collection>ProQuest Central UK/Ireland</collection><collection>ProQuest Central Essentials</collection><collection>ProQuest Central</collection><collection>Technology Collection</collection><collection>ProQuest One Community College</collection><collection>ProQuest Central Korea</collection><collection>SciTech Premium Collection</collection><collection>ProQuest Engineering Collection</collection><collection>Engineering Database</collection><collection>Publicly Available Content Database</collection><collection>ProQuest One Academic Eastern Edition (DO NOT USE)</collection><collection>ProQuest One Academic</collection><collection>ProQuest One Academic UKI Edition</collection><collection>ProQuest Central China</collection><collection>Engineering Collection</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext</fulltext></delivery><addata><au>Sah, Sudhakar</au><au>Kumar, Ravish</au><au>Rohmetra, Honnesh</au><au>Saboori, Ehsan</au><format>book</format><genre>document</genre><ristype>GEN</ristype><atitle>Token Pruning using a Lightweight Background Aware Vision Transformer</atitle><jtitle>arXiv.org</jtitle><date>2024-10-12</date><risdate>2024</risdate><eissn>2331-8422</eissn><abstract>High runtime memory and high latency puts significant constraint on Vision Transformer training and inference, especially on edge devices. Token pruning reduces the number of input tokens to the ViT based on importance criteria of each token. We present a Background Aware Vision Transformer (BAViT) model, a pre-processing block to object detection models like DETR/YOLOS aimed to reduce runtime memory and increase throughput by using a novel approach to identify background tokens in the image. The background tokens can be pruned completely or partially before feeding to a ViT based object detector. We use the semantic information provided by segmentation map and/or bounding box annotation to train a few layers of ViT to classify tokens to either foreground or background. Using 2 layers and 10 layers of BAViT, background and foreground tokens can be separated with 75% and 88% accuracy on VOC dataset and 71% and 80% accuracy on COCO dataset respectively. We show a 2 layer BAViT-small model as pre-processor to YOLOS can increase the throughput by 30% - 40% with a mAP drop of 3% without any sparse fine-tuning and 2% with sparse fine-tuning. Our approach is specifically targeted for Edge AI use cases.</abstract><cop>Ithaca</cop><pub>Cornell University Library, arXiv.org</pub><oa>free_for_read</oa></addata></record>
fulltext fulltext
identifier EISSN: 2331-8422
ispartof arXiv.org, 2024-10
issn 2331-8422
language eng
recordid cdi_proquest_journals_3116744293
source Free E- Journals
subjects Accuracy
Annotations
Datasets
Image segmentation
Memory devices
Microprocessors
Object recognition
Pruning
title Token Pruning using a Lightweight Background Aware Vision Transformer
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-25T21%3A13%3A10IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-proquest&rft_val_fmt=info:ofi/fmt:kev:mtx:book&rft.genre=document&rft.atitle=Token%20Pruning%20using%20a%20Lightweight%20Background%20Aware%20Vision%20Transformer&rft.jtitle=arXiv.org&rft.au=Sah,%20Sudhakar&rft.date=2024-10-12&rft.eissn=2331-8422&rft_id=info:doi/&rft_dat=%3Cproquest%3E3116744293%3C/proquest%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=3116744293&rft_id=info:pmid/&rfr_iscdi=true