What Makes for Good Visual Instructions? Synthesizing Complex Visual Reasoning Instructions for Visual Instruction Tuning
Visual instruction tuning is an essential approach to improving the zero-shot generalization capability of Multi-modal Large Language Models (MLLMs). A surge of visual instruction datasets with various focuses and characteristics have been proposed recently, enabling MLLMs to achieve surprising resu...
Gespeichert in:
Veröffentlicht in: | arXiv.org 2023-11 |
---|---|
Hauptverfasser: | , , , , , , , , |
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 | Du, Yifan Guo, Hangyu Zhou, Kun Wayne Xin Zhao Wang, Jinpeng Wang, Chuyuan Cai, Mingchen Song, Ruihua Ji-Rong, Wen |
description | Visual instruction tuning is an essential approach to improving the zero-shot generalization capability of Multi-modal Large Language Models (MLLMs). A surge of visual instruction datasets with various focuses and characteristics have been proposed recently, enabling MLLMs to achieve surprising results on evaluation benchmarks. To develop more capable MLLMs, in this paper, we aim to investigate a more fundamental question: ``what makes for good visual instructions?''. By conducting a comprehensive empirical study, we find that instructions focused on complex visual reasoning tasks are particularly effective in improving the performance of MLLMs on evaluation benchmarks. Building upon this finding, we design a systematic approach to automatically creating high-quality complex visual reasoning instructions. Our approach employs a synthesis-complication-reformulation paradigm, leveraging multiple stages to gradually increase the complexity of the instructions while guaranteeing quality. Based on this approach, we create the synthetic visual reasoning instruction dataset consisting of 32K examples, namely ComVint, and fine-tune four MLLMs on it. Experimental results demonstrate that our dataset consistently enhances the performance of all the compared MLLMs, e.g., improving the performance of MiniGPT-4 and BLIP-2 on MME-Cognition by 32.6% and 28.8%, respectively. Our code and data are publicly available at the link: https://github.com/RUCAIBox/ComVint. |
format | Article |
fullrecord | <record><control><sourceid>proquest</sourceid><recordid>TN_cdi_proquest_journals_2886463835</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2886463835</sourcerecordid><originalsourceid>FETCH-proquest_journals_28864638353</originalsourceid><addsrcrecordid>eNqNjL0KwjAYRYMgWLTvEHAu1KSt2RyKf4OLio4laGpTa1LzJWB9eq0oCC5Odzjnng7yCKWjgEWE9JAPUIZhSJIxiWPqoWZfcItX_CwA59rgudZHvJPgeIWXCqxxByu1ggneNMoWAuRdqhNO9aWuxO1jrgUHrVrw_XkFf1t461p1gLo5r0D47-2j4Wy6TRdBbfTVCbBZqZ1RT5QRxpIooYzG9D_rATALTy0</addsrcrecordid><sourcetype>Aggregation Database</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype><pqid>2886463835</pqid></control><display><type>article</type><title>What Makes for Good Visual Instructions? Synthesizing Complex Visual Reasoning Instructions for Visual Instruction Tuning</title><source>Free E- Journals</source><creator>Du, Yifan ; Guo, Hangyu ; Zhou, Kun ; Wayne Xin Zhao ; Wang, Jinpeng ; Wang, Chuyuan ; Cai, Mingchen ; Song, Ruihua ; Ji-Rong, Wen</creator><creatorcontrib>Du, Yifan ; Guo, Hangyu ; Zhou, Kun ; Wayne Xin Zhao ; Wang, Jinpeng ; Wang, Chuyuan ; Cai, Mingchen ; Song, Ruihua ; Ji-Rong, Wen</creatorcontrib><description>Visual instruction tuning is an essential approach to improving the zero-shot generalization capability of Multi-modal Large Language Models (MLLMs). A surge of visual instruction datasets with various focuses and characteristics have been proposed recently, enabling MLLMs to achieve surprising results on evaluation benchmarks. To develop more capable MLLMs, in this paper, we aim to investigate a more fundamental question: ``what makes for good visual instructions?''. By conducting a comprehensive empirical study, we find that instructions focused on complex visual reasoning tasks are particularly effective in improving the performance of MLLMs on evaluation benchmarks. Building upon this finding, we design a systematic approach to automatically creating high-quality complex visual reasoning instructions. Our approach employs a synthesis-complication-reformulation paradigm, leveraging multiple stages to gradually increase the complexity of the instructions while guaranteeing quality. Based on this approach, we create the synthetic visual reasoning instruction dataset consisting of 32K examples, namely ComVint, and fine-tune four MLLMs on it. Experimental results demonstrate that our dataset consistently enhances the performance of all the compared MLLMs, e.g., improving the performance of MiniGPT-4 and BLIP-2 on MME-Cognition by 32.6% and 28.8%, respectively. Our code and data are publicly available at the link: https://github.com/RUCAIBox/ComVint.</description><identifier>EISSN: 2331-8422</identifier><language>eng</language><publisher>Ithaca: Cornell University Library, arXiv.org</publisher><subject>Benchmarks ; Cognition ; Datasets ; Large language models ; Performance evaluation ; Reasoning ; Task complexity ; Tuning ; Visual tasks</subject><ispartof>arXiv.org, 2023-11</ispartof><rights>2023. This work is published under http://arxiv.org/licenses/nonexclusive-distrib/1.0/ (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>Du, Yifan</creatorcontrib><creatorcontrib>Guo, Hangyu</creatorcontrib><creatorcontrib>Zhou, Kun</creatorcontrib><creatorcontrib>Wayne Xin Zhao</creatorcontrib><creatorcontrib>Wang, Jinpeng</creatorcontrib><creatorcontrib>Wang, Chuyuan</creatorcontrib><creatorcontrib>Cai, Mingchen</creatorcontrib><creatorcontrib>Song, Ruihua</creatorcontrib><creatorcontrib>Ji-Rong, Wen</creatorcontrib><title>What Makes for Good Visual Instructions? Synthesizing Complex Visual Reasoning Instructions for Visual Instruction Tuning</title><title>arXiv.org</title><description>Visual instruction tuning is an essential approach to improving the zero-shot generalization capability of Multi-modal Large Language Models (MLLMs). A surge of visual instruction datasets with various focuses and characteristics have been proposed recently, enabling MLLMs to achieve surprising results on evaluation benchmarks. To develop more capable MLLMs, in this paper, we aim to investigate a more fundamental question: ``what makes for good visual instructions?''. By conducting a comprehensive empirical study, we find that instructions focused on complex visual reasoning tasks are particularly effective in improving the performance of MLLMs on evaluation benchmarks. Building upon this finding, we design a systematic approach to automatically creating high-quality complex visual reasoning instructions. Our approach employs a synthesis-complication-reformulation paradigm, leveraging multiple stages to gradually increase the complexity of the instructions while guaranteeing quality. Based on this approach, we create the synthetic visual reasoning instruction dataset consisting of 32K examples, namely ComVint, and fine-tune four MLLMs on it. Experimental results demonstrate that our dataset consistently enhances the performance of all the compared MLLMs, e.g., improving the performance of MiniGPT-4 and BLIP-2 on MME-Cognition by 32.6% and 28.8%, respectively. Our code and data are publicly available at the link: https://github.com/RUCAIBox/ComVint.</description><subject>Benchmarks</subject><subject>Cognition</subject><subject>Datasets</subject><subject>Large language models</subject><subject>Performance evaluation</subject><subject>Reasoning</subject><subject>Task complexity</subject><subject>Tuning</subject><subject>Visual tasks</subject><issn>2331-8422</issn><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2023</creationdate><recordtype>article</recordtype><sourceid>BENPR</sourceid><recordid>eNqNjL0KwjAYRYMgWLTvEHAu1KSt2RyKf4OLio4laGpTa1LzJWB9eq0oCC5Odzjnng7yCKWjgEWE9JAPUIZhSJIxiWPqoWZfcItX_CwA59rgudZHvJPgeIWXCqxxByu1ggneNMoWAuRdqhNO9aWuxO1jrgUHrVrw_XkFf1t461p1gLo5r0D47-2j4Wy6TRdBbfTVCbBZqZ1RT5QRxpIooYzG9D_rATALTy0</recordid><startdate>20231102</startdate><enddate>20231102</enddate><creator>Du, Yifan</creator><creator>Guo, Hangyu</creator><creator>Zhou, Kun</creator><creator>Wayne Xin Zhao</creator><creator>Wang, Jinpeng</creator><creator>Wang, Chuyuan</creator><creator>Cai, Mingchen</creator><creator>Song, Ruihua</creator><creator>Ji-Rong, Wen</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>20231102</creationdate><title>What Makes for Good Visual Instructions? Synthesizing Complex Visual Reasoning Instructions for Visual Instruction Tuning</title><author>Du, Yifan ; Guo, Hangyu ; Zhou, Kun ; Wayne Xin Zhao ; Wang, Jinpeng ; Wang, Chuyuan ; Cai, Mingchen ; Song, Ruihua ; Ji-Rong, Wen</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-proquest_journals_28864638353</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2023</creationdate><topic>Benchmarks</topic><topic>Cognition</topic><topic>Datasets</topic><topic>Large language models</topic><topic>Performance evaluation</topic><topic>Reasoning</topic><topic>Task complexity</topic><topic>Tuning</topic><topic>Visual tasks</topic><toplevel>online_resources</toplevel><creatorcontrib>Du, Yifan</creatorcontrib><creatorcontrib>Guo, Hangyu</creatorcontrib><creatorcontrib>Zhou, Kun</creatorcontrib><creatorcontrib>Wayne Xin Zhao</creatorcontrib><creatorcontrib>Wang, Jinpeng</creatorcontrib><creatorcontrib>Wang, Chuyuan</creatorcontrib><creatorcontrib>Cai, Mingchen</creatorcontrib><creatorcontrib>Song, Ruihua</creatorcontrib><creatorcontrib>Ji-Rong, Wen</creatorcontrib><collection>ProQuest SciTech Collection</collection><collection>ProQuest Technology Collection</collection><collection>Materials Science & 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>Du, Yifan</au><au>Guo, Hangyu</au><au>Zhou, Kun</au><au>Wayne Xin Zhao</au><au>Wang, Jinpeng</au><au>Wang, Chuyuan</au><au>Cai, Mingchen</au><au>Song, Ruihua</au><au>Ji-Rong, Wen</au><format>book</format><genre>document</genre><ristype>GEN</ristype><atitle>What Makes for Good Visual Instructions? Synthesizing Complex Visual Reasoning Instructions for Visual Instruction Tuning</atitle><jtitle>arXiv.org</jtitle><date>2023-11-02</date><risdate>2023</risdate><eissn>2331-8422</eissn><abstract>Visual instruction tuning is an essential approach to improving the zero-shot generalization capability of Multi-modal Large Language Models (MLLMs). A surge of visual instruction datasets with various focuses and characteristics have been proposed recently, enabling MLLMs to achieve surprising results on evaluation benchmarks. To develop more capable MLLMs, in this paper, we aim to investigate a more fundamental question: ``what makes for good visual instructions?''. By conducting a comprehensive empirical study, we find that instructions focused on complex visual reasoning tasks are particularly effective in improving the performance of MLLMs on evaluation benchmarks. Building upon this finding, we design a systematic approach to automatically creating high-quality complex visual reasoning instructions. Our approach employs a synthesis-complication-reformulation paradigm, leveraging multiple stages to gradually increase the complexity of the instructions while guaranteeing quality. Based on this approach, we create the synthetic visual reasoning instruction dataset consisting of 32K examples, namely ComVint, and fine-tune four MLLMs on it. Experimental results demonstrate that our dataset consistently enhances the performance of all the compared MLLMs, e.g., improving the performance of MiniGPT-4 and BLIP-2 on MME-Cognition by 32.6% and 28.8%, respectively. Our code and data are publicly available at the link: https://github.com/RUCAIBox/ComVint.</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, 2023-11 |
issn | 2331-8422 |
language | eng |
recordid | cdi_proquest_journals_2886463835 |
source | Free E- Journals |
subjects | Benchmarks Cognition Datasets Large language models Performance evaluation Reasoning Task complexity Tuning Visual tasks |
title | What Makes for Good Visual Instructions? Synthesizing Complex Visual Reasoning Instructions for Visual Instruction Tuning |
url | https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-01-28T15%3A05%3A24IST&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=What%20Makes%20for%20Good%20Visual%20Instructions?%20Synthesizing%20Complex%20Visual%20Reasoning%20Instructions%20for%20Visual%20Instruction%20Tuning&rft.jtitle=arXiv.org&rft.au=Du,%20Yifan&rft.date=2023-11-02&rft.eissn=2331-8422&rft_id=info:doi/&rft_dat=%3Cproquest%3E2886463835%3C/proquest%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_pqid=2886463835&rft_id=info:pmid/&rfr_iscdi=true |