[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-83443-en":3,"doc-seo-83443-105":30,"detail-sidebar-cat-0-en-105":92},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":21,"is_downloadable":21,"audit_status":21,"page_count":22,"language":23,"language_code":24,"site_id":25,"html_lang":24,"table_of_contents":26,"faqs":27,"seo_title":13,"seo_description":14,"update_tm":28,"read_time":29},83443,1099513958607,"Jiven","https://ap-avatar.wpscdn.com/avatar/100002390cf8733938c?x-image-process=image/resize,m_fixed,w_180,h_180&k=1778829742770036399",8,"Research & Report","Decompose, Compare, and Decide: Multimodal LLMs are Implicit Few-Shot Learners","Multimodal Large Language Models (MLLMs) excel at image analysis, yet few-shot image classification with limited labeled examples remains difficult. DeCoDe introduces a simple decomposition method that turns off-the-shelf MLLMs into strong few-shot classifiers without additional training. The approach reformulates few-shot classification as pairwise support–query comparisons, prompting the model to decide whether two images belong to the same class and using the “Yes” logit as a similarity score for class ranking. Adding high-level domain information further improves results. Extensive evaluation across twelve diverse datasets shows significant gains over state-of-the-art few-shot methods on both standard and newly curated domains.","arXiv :2607 .00125v1 [ cs .CV] 30 Jun 2026  \nDecompose, Compare, and Decide: Multimodal LLMs are Implicit Few-Shot Learners  \nYunhan Wang, Eshika Khandelwal, Edson Araujo, Walid Bousselham, Nina Shvetsova, and Hilde Kuehne  \nTuebingen AI Center, University of Tuebingen, Germany [yunhan.wang@student.uni-tuebingen.de](yunhan.wang@student.uni-tuebingen.de)  \nAbstract. Multimodal Large Language Models (MLLMs) have demonstrated remarkable abilities when analyzing images, yet translating these capabilities to few-shot image classification remains challenging. To bridge this gap, we present DeCoDe, a simple yet effective technique that enables off-the-shelf MLLMs to act as strong few-shot classifiers without any additional training. Our approach builds on the idea of few-shot classification as a set of pairwise image comparisons, decomposing the task into a set of binary decisions. Given a query image and a support image from a candidate class, the MLLM is prompted to decide whether the two images depict the same class. The logit corresponding to an affirmative response is then used as a similarity score to assign the query image to the most likely class. While this already yields good results, we show that providing additional high-level information, such as the data domain, to the model further improves performance. Our evaluation provides an extensive analysis of various inference variants on a suite of twelve datasets, six established and six newly curated few-shot benchmarks spanning across diverse domains. The results show that the proposed simple decomposition technique can turn off-the-shelf MLLMs into powerful few-shot learners, significantly outperforming current stateof-the-art few-shot methods on both standard and novel domains. Code is available at [https://github.com/yunhanwang1105/DeCoDe](https://github.com/yunhanwang1105/DeCoDe).  \nKeywords: Few-Shot Learning · Multimodal Large Language Models · Vision-Language Models · In-Context Learning  \n1 Introduction  \nFew-shot learning (FSL) enables models to generalize to new tasks and domains using only a small number of labeled examples per class. This can be especially valuable for real-world scenarios where large annotated datasets are unavailable or adaptation to novel tasks is required at inference time. Early few-shot methods required model training on a set of base classes in order to enable recognition of held-out novel classes [12,21,34,39,41,44] . With the emergence of foundation models trained on large-scale web data via self-or weakly-supervised objectives, such as DINO [30] or CLIP [33], the focus shifted towards exploring  \n2 Y. Wang et al  \nIn-context Prompting  \n.  \nWhat is this? Class 2.  \nWhat is this? Class 3.  \nSo what is this? Choose from …  \nimg token  \n……  \ntext token  \nMLLM  \nOutput: “Class 1.”   \nDecomposed Prompting  \nFig. 1: We propose a decomposed prompting technique (DeCoDe) for few-shot classification with MLLMs. We decompose the task into pairwise support–query comparisons, asking whether two images belong to the same class. By ranking the model’s affirmative responses across candidate pairs (compare) and selecting the highest-scoring logit as the predicted class (decide), MLLMs become strong few-shot classifiers without any training. Unlike standard in-context prompting, which relies heavily on semantic label names, our decomposed approach succeeds even with anonymized labels by forcing the model to perform direct visual-to-visual comparison.  \nhow pretrained representations can be leveraged for improved few-shot classification with minimal or no additional training [6,8,14,42,50,52,53] . More recently, multimodal large language models (MLLMs), such as LLaVA-OneVision [22], InternVL3 [55], or Qwen3-VL [3], have demonstrated strong vision-language capabilities, raising the question of whether such models can also serve as effective few-shot learners [26, 46] .  \nSo far, MLLMs have been explored for few-shot learning in three ways: by extracting ","cbCaikXrW6JlCsh4","https://ap.wps.com/l/cbCaikXrW6JlCsh4","pdf",3016113,7,1,31,"English","en",105,"# Introduction\n## In-context Prompting\n## Decomposed Prompting (DeCoDe)\n## Pairwise “Decompose, Compare, Decide” Workflow","[{\"question\":\"What problem does DeCoDe address for multimodal LLMs?\",\"answer\":\"DeCoDe targets the gap between strong MLLM image understanding and the difficulty of performing few-shot image classification effectively with few labeled examples.\"},{\"question\":\"How does DeCoDe convert few-shot classification into model prompts?\",\"answer\":\"It decomposes the task into binary pairwise prompts, asking whether a query image and a support image depict the same class, then ranks candidate classes using the affirmative (“Yes”) logit.\"},{\"question\":\"Why can providing domain information improve DeCoDe’s performance?\",\"answer\":\"The method shows that contextual high-level information about the data domain (often derived from the dataset name) helps the model make better pairwise decisions, improving overall few-shot accuracy.\"}]",1784187793,78,{"code":4,"msg":31,"data":32},"ok",{"site_id":25,"language":24,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":87,"head_meta":89,"extra_data":91,"updated_unix":28},"decompose-compare-and-decide-multimodal-llms-are-implicit-few-shot-learners","",{"@graph":36,"@context":86},[37,54,69],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":21},"https://docshare.wps.com","Home","ListItem",{"item":45,"name":46,"@type":43,"position":47},"https://docshare.wps.com/document/","Document",2,{"item":49,"name":12,"@type":43,"position":50},"https://docshare.wps.com/document/research-report/",3,{"item":52,"name":13,"@type":43,"position":53},"https://docshare.wps.com/document/decompose-compare-and-decide-multimodal-llms-are-implicit-few-shot-learners/83443/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":24,"description":14,"dateModified":62,"datePublished":63,"encodingFormat":61,"isAccessibleForFree":64,"interactionStatistic":65},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":41,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-07-25","2026-07-16",true,{"@type":66,"interactionType":67,"userInteractionCount":20},"InteractionCounter",{"@type":68},"ViewAction",{"@type":70,"mainEntity":71},"FAQPage",[72,78,82],{"name":73,"@type":74,"acceptedAnswer":75},"What problem does DeCoDe address for multimodal LLMs?","Question",{"text":76,"@type":77},"DeCoDe targets the gap between strong MLLM image understanding and the difficulty of performing few-shot image classification effectively with few labeled examples.","Answer",{"name":79,"@type":74,"acceptedAnswer":80},"How does DeCoDe convert few-shot classification into model prompts?",{"text":81,"@type":77},"It decomposes the task into binary pairwise prompts, asking whether a query image and a support image depict the same class, then ranks candidate classes using the affirmative (“Yes”) logit.",{"name":83,"@type":74,"acceptedAnswer":84},"Why can providing domain information improve DeCoDe’s performance?",{"text":85,"@type":77},"The method shows that contextual high-level information about the data domain (often derived from the dataset name) helps the model make better pairwise decisions, improving overall few-shot accuracy.","https://schema.org",{"og:url":52,"og:type":88,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":90,"canonical":52},"index,follow",{"doc_id":7,"site_id":25},{"code":4,"msg":5,"data":93},[94,98,102,106,111,116,120,123,128,131,135],{"id":21,"doc_module":4,"doc_module_name":46,"category_name":95,"show_sort_weight":96,"slug":97},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":99,"show_sort_weight":100,"slug":101},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":46,"category_name":103,"show_sort_weight":104,"slug":105},"Exam",70,"exam",{"id":107,"doc_module":4,"doc_module_name":46,"category_name":108,"show_sort_weight":109,"slug":110},5,"Comic",60,"comic",{"id":112,"doc_module":4,"doc_module_name":46,"category_name":113,"show_sort_weight":114,"slug":115},6,"Technology",50,"technology",{"id":20,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":46,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":46,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":46,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":46,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":46,"category_name":137,"show_sort_weight":107,"slug":138},19,"General","general"]