[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-122379-en":3,"doc-seo-122379-105":30,"detail-sidebar-cat-0-en-105":91},{"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":4,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":27,"seo_description":14,"update_tm":28,"read_time":29},122379,962075114765,"Quinn","https://ap-avatar.wpscdn.com/davatar_a8503ba1806abce46bf441b54a3ca4cd",8,"Research & Report","Optimized Visual Anomaly detection models for tiny machine learning","This thesis examines visual anomaly detection in the setting of Tiny Machine Learning (TinyML), targeting optimized deployment on resource-constrained edge devices. Several state-of-the-art unsupervised VAD models, including PaDiM and PatchCore, are benchmarked on MVTec, ViSA, and Real-IAD to establish performance baselines. The work further analyzes model behavior under noisy or contaminated training data. Targeted optimizations are then introduced to increase edge efficiency, such as simplifying PaDiM distance computations and using quantization in PatchCore to reduce memory usage.","University of Padova  \nDepartment of Information Engineering Master Thesis in Computer Engineering  \nOptimized Visual Anomaly detection models for tiny machine learning  \nSupervisor Master Candidate  \nProf. Gian Antonio Susto Youssef Ben Khalifa  \nUniversity of Padova  \nCo-supervisor  \nDr. Davide Dalle Pezze Phd. Manuel Barusco  \nPhd. Francesco Borsatti University of Padova  \nAcademic Year  \n2024-2025  \nii  \nLearn from yesterday, live for today, hope for tomorrow. The important thing is not to stop questioning.  \nAlbert Einstein  \niv  \nAbstract  \nThis thesis explores the field of visual anomaly detection within the context of Tiny Machine Learning (TinyML), focusing on the optimization of existing models for deployment on resourceconstrained edge devices. We begin by benchmarking several state-of-the-art unsupervised VAD models,such as PaDiM and PatchCore,on popular datasets like MVTec, ViSA, and Real-IAD. These benchmarks provide a performance baseline for comparison. We then explore how these models behave when trained on noisy or contaminated data. Finally, we apply specific optimizations to improve the models’ performance on edge devices. These include simplifying distance calculations for PaDiM and applying quantization techniques to reduce memory usage in PatchCore. Our results show that it is possible to significantly improve the efficiency of these models without heavily impacting the performance of the original models, making them more suitable for real-world deployment on low-resource devices.  \nvi  \nContents  \nAbstract v  \nList of figures ix  \nList of tables xi  \nListing of acronyms xiii  \n1 Introduction 1  \n2 Related Work 3  \n2.1 VAD Models ................................. 4  \n2.2 PaSTE ..................................... 5  \n2.3 Product Quantization ............................. 5  \n2.4 Benchmark Datasets .............................. 6  \n3 Benchmark Baseline for IAD Datasets and Models 7  \n3.1 MVTec Dataset ................................ 7  \n3.2 Real-IAD Dataset ............................... 8  \n3.3 ViSA Dataset ................................. 11  \n3.4 Feature Extraction Backbones ......................... 11  \n3.5 Benchmark Setup ............................... 13  \n3.6 Performance Metrics ............................. 14  \n3.7 Benchmark Results .............................. 15  \n3.8 Observations ................................. 15  \n4 VAD Model in a Noisy Scenario 21  \n4.1 Training set contamination .......................... 21  \n4.2 Contaminated Benchmark .......................... 23  \n4.3 Contaminated Benchmark Results ...................... 23  \n5 PaDiM Optimizations 25  \n5.1 PaDiM Model ................................. 25  \n5.2 Mahalanobis Distance With Diagonal Covariance .............. 27  \n6 PatchCore Optimizations 29  \n6.1 PatchCore Model ............................... 29  \n6.2 Coreset Greedy Selection ........................... 30  \n6.3 Memory-bank Quantization ......................... 31  \n6.3.1 Product Quantization ........................ 32  \n6.3.2 Partially Quantized Nearest Neighbor Search ............ 32  \n6.3.3 Hyperparameter Tuning: K and c in PQNN-Based Anomaly Detection .................................. 34  \n7 Experiments 37  \n7.1 Experimental Setup .............................. 38  \n7.2 MoViAD Library ............................... 38  \n8 Experiment Results 39  \n8.1 Evaluation metrics ............................... 39  \n8.2 PaDiM Experiments .............................. 40  \n8.3 PatchCore Experiments ............................ 41  \n8.3.1 Quantized PatchCore ......................... 41  \n8.3.2 PQNN Search Performance ..................... 42  \n9 Conclusions & Future Work 45  \n9.1 Experiment Results Observations ....................... 45  \n9.2 Final Considerations and Conclusions .................... 45  \n9.3 Future Work .................................. 46  \n10 Appendix 49  \nReferences 59  \nAcknowledgments 63  \nListing of figures  \n3.1 MVTec Dataset Entries Examp","cbCaieeuE8e0fAz8","https://ap.wps.com/l/cbCaieeuE8e0fAz8","pdf",10321303,1,77,"English","en",105,"# Abstract\n# 1 Introduction\n# 2 Related Work\n## 2.1 VAD Models\n## 2.2 PaSTE\n## 2.3 Product Quantization\n## 2.4 Benchmark Datasets\n# 3 Benchmark Baseline for IAD Datasets and Models\n## 3.1 MVTec Dataset\n## 3.2 Real-IAD Dataset\n## 3.3 ViSA Dataset\n## 3.4 Feature Extraction Backbones\n## 3.5 Benchmark Setup\n## 3.6 Performance Metrics\n## 3.7 Benchmark Results\n## 3.8 Observations\n# 4 VAD Model in a Noisy Scenario\n## 4.1 Training set contamination\n## 4.2 Contaminated Benchmark\n## 4.3 Contaminated Benchmark Results\n# 5 PaDiM Optimizations\n## 5.1 PaDiM Model\n## 5.2 Mahalanobis Distance With Diagonal Covariance\n# 6 PatchCore Optimizations\n## 6.1 PatchCore Model\n## 6.2 Coreset Greedy Selection\n## 6.3 Memory-bank Quantization\n# 7 Experiments\n## 7.1 Experimental Setup\n## 7.2 MoViAD Library\n# 8 Experiment Results\n## 8.1 Evaluation metrics\n## 8.2 PaDiM Experiments\n## 8.3 PatchCore Experiments\n# 9 Conclusions & Future Work\n## 9.1 Experiment Results Observations\n## 9.2 Final Considerations and Conclusions\n## 9.3 Future Work\n# 10 Appendix\n# References\n# Acknowledgments","[{\"question\":\"What is the thesis scope within Tiny Machine Learning (TinyML)?\",\"answer\":\"The thesis studies visual anomaly detection for deployment on resource-constrained edge devices, emphasizing optimization of existing unsupervised models for TinyML scenarios.\"},{\"question\":\"How are baseline VAD models evaluated in the thesis?\",\"answer\":\"PaDiM and PatchCore are benchmarked on datasets such as MVTec, ViSA, and Real-IAD to provide a performance baseline for later comparisons.\"},{\"question\":\"What optimizations are proposed for edge-device efficiency?\",\"answer\":\"The work simplifies distance calculations for PaDiM and applies quantization techniques for PatchCore to reduce memory usage, aiming to improve efficiency without strongly degrading original performance.\"}]","Optimized Visual Anomaly detection models for tiny machine learning | PDF",1785810332,194,{"code":4,"msg":31,"data":32},"ok",{"site_id":24,"language":23,"slug":33,"title":13,"keywords":34,"description":14,"schema_data":35,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":28},"optimized-visual-anomaly-detection-models-for-tiny-machine-learning","",{"@graph":36,"@context":85},[37,54,68],{"@type":38,"itemListElement":39},"BreadcrumbList",[40,44,48,51],{"item":41,"name":42,"@type":43,"position":20},"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/optimized-visual-anomaly-detection-models-for-tiny-machine-learning/122379/",4,{"url":52,"name":13,"@type":55,"author":56,"headline":13,"publisher":58,"fileFormat":61,"inLanguage":23,"description":14,"dateModified":62,"datePublished":62,"encodingFormat":61,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":57},"Person",{"url":41,"name":59,"@type":60},"DocShare","Organization","application/pdf","2026-08-04",true,{"@type":65,"interactionType":66,"userInteractionCount":4},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What is the thesis scope within Tiny Machine Learning (TinyML)?","Question",{"text":75,"@type":76},"The thesis studies visual anomaly detection for deployment on resource-constrained edge devices, emphasizing optimization of existing unsupervised models for TinyML scenarios.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How are baseline VAD models evaluated in the thesis?",{"text":80,"@type":76},"PaDiM and PatchCore are benchmarked on datasets such as MVTec, ViSA, and Real-IAD to provide a performance baseline for later comparisons.",{"name":82,"@type":73,"acceptedAnswer":83},"What optimizations are proposed for edge-device efficiency?",{"text":84,"@type":76},"The work simplifies distance calculations for PaDiM and applies quantization techniques for PatchCore to reduce memory usage, aiming to improve efficiency without strongly degrading original performance.","https://schema.org",{"og:url":52,"og:type":87,"og:title":13,"og:site_name":59,"og:description":14},"article",{"robots":89,"canonical":52},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":20,"doc_module":4,"doc_module_name":46,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":47,"doc_module":4,"doc_module_name":46,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":53,"doc_module":4,"doc_module_name":46,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":46,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":46,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":46,"category_name":117,"show_sort_weight":118,"slug":119},7,"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":106,"slug":138},19,"General","general"]