{"id":{"repo_id":"usf","oai_identifier":"oai:digitalcommons.usf.edu:etd-12190"},"canonical_url":"https://search.ndltd.org/etd/usf/oai:digitalcommons.usf.edu:etd-12190","repository":{"repo_id":"usf","name":"University of South Florida","base_url":"https://digitalcommons.usf.edu/do/oai/"},"display":{"title":"EDGE-STREAMER: A Lightweight, Self-Supervised Event Segmentation Model","abstract":"Event segmentation is the practice of autonomously detecting the boundaries of semantically connected sequences of actions within a video. It is connected to many areas of computer vision, including action recognition and event understanding. Nearly all current work requires the entire video to be stored in memory to be processed in multiple passes. This takes up valuable resources, increases processing time, prevents the use of low-power, low-memory devices, and precludes long-form content and live videos from undergoing event segmentation. This thesis introduces EDGE-STREAMER, a real-time, lightweight, self-supervised, transformer architecture capable of performing event segmentation in a single pass. EDGE-STREAMER uses techniques from reinforcement learning models to store representative information about the past to predict events, rather than the entire video. It takes advantage of weight sharing to reduce model size, further reducing its memory footprint and its training requirements. During training the model processes unlabeled video data in a single pass, producing event boundaries in real time. On Epic-Kitchens EDGE-STREAMER achieves the same MoF performance as STREAMER within 2\\%. On an NVIDIA Jetson Orin Nano, it sustains over 16 fps, showing that it is capable of real-time performance on an edge device. The contributions of this thesis are 1. Four low-memory event segmentation architectures. 2. A thorough analysis comparing the accuracies of EDGE-STREAMER to the state-of-the-art. 3. Experimental data showing the viability of EDGE-STREAMER on edge hardware. These results open the door for embedded device-based cameras, such as those used for surveillance, autonomous drones, and wearables, and provide possibilities for future work, including multi-modal streaming and embedded-device continual learning.","abstract_html":"Event segmentation is the practice of autonomously detecting the boundaries of semantically connected sequences of actions within a video. It is connected to many areas of computer vision, including action recognition and event understanding. Nearly all current work requires the entire video to be stored in memory to be processed in multiple passes. This takes up valuable resources, increases processing time, prevents the use of low-power, low-memory devices, and precludes long-form content and live videos from undergoing event segmentation. This thesis introduces EDGE-STREAMER, a real-time, lightweight, self-supervised, transformer architecture capable of performing event segmentation in a single pass. EDGE-STREAMER uses techniques from reinforcement learning models to store representative information about the past to predict events, rather than the entire video. It takes advantage of weight sharing to reduce model size, further reducing its memory footprint and its training requirements. During training the model processes unlabeled video data in a single pass, producing event boundaries in real time. On Epic-Kitchens EDGE-STREAMER achieves the same MoF performance as STREAMER within 2\\%. On an NVIDIA Jetson Orin Nano, it sustains over 16 fps, showing that it is capable of real-time performance on an edge device. The contributions of this thesis are 1. Four low-memory event segmentation architectures. 2. A thorough analysis comparing the accuracies of EDGE-STREAMER to the state-of-the-art. 3. Experimental data showing the viability of EDGE-STREAMER on edge hardware. These results open the door for embedded device-based cameras, such as those used for surveillance, autonomous drones, and wearables, and provide possibilities for future work, including multi-modal streaming and embedded-device continual learning.","abstract_has_math":false,"creators":["Miller, Lucas"],"institution":"University of South Florida","degree_name":"MS in Computer Engineering (M.S.C.P.)","degree_level":"Thesis","degree_discipline":"Computer Science and Engineering","degree_department":null,"school":null,"contributors":["Sudeep Sarkar, Ph.D.","Shaun Canavan, Ph.D.","Mauricio Pamplona Segundo, Ph.D."],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025-07-08T07:00:00Z","date_published":"2025-07-08T07:00:00Z","updated_at":"2026-11-19T08:49:48Z","subjects":["Transformers","Online Machine Learning","Edge Computing","Event Understanding","Representation Learning","Computer Engineering"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[],"person_identifier_entries":[]},"links":{"outbound_url":"https://digitalcommons.usf.edu/etd/10887","outbound_label":"Repository record","outbound_source":"dc:identifier","record_url":"https://digitalcommons.usf.edu/etd/10887","record_label":"Repository record","record_kind":"record","record_source":"dc:identifier","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://digitalcommons.usf.edu/do/oai/?verb=GetRecord&metadataPrefix=oai_etdms&identifier=oai%3Adigitalcommons.usf.edu%3Aetd-12190","prefix":"oai_etdms"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Sudeep Sarkar, Ph.D.","Shaun Canavan, Ph.D.","Mauricio Pamplona Segundo, Ph.D."]},{"key":"dc:creator","label":"Author","values":["Miller, Lucas"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2025-07-08T07:00:00Z"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Computer Science and Engineering"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Thesis","M.S.Cp."]},{"key":"thesis:degree_name","label":"Degree Name","values":["MS in Computer Engineering (M.S.C.P.)"]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["University of South Florida"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Transformers","Online Machine Learning","Edge Computing","Event Understanding","Representation Learning","Computer Engineering"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://digitalcommons.usf.edu/etd/10887"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Event segmentation is the practice of autonomously detecting the boundaries of semantically connected sequences of actions within a video. It is connected to many areas of computer vision, including action recognition and event understanding. Nearly all current work requires the entire video to be stored in memory to be processed in multiple passes. This takes up valuable resources, increases processing time, prevents the use of low-power, low-memory devices, and precludes long-form content and live videos from undergoing event segmentation. This thesis introduces EDGE-STREAMER, a real-time, lightweight, self-supervised, transformer architecture capable of performing event segmentation in a single pass. EDGE-STREAMER uses techniques from reinforcement learning models to store representative information about the past to predict events, rather than the entire video. It takes advantage of weight sharing to reduce model size, further reducing its memory footprint and its training requirements. During training the model processes unlabeled video data in a single pass, producing event boundaries in real time. On Epic-Kitchens EDGE-STREAMER achieves the same MoF performance as STREAMER within 2\\%. On an NVIDIA Jetson Orin Nano, it sustains over 16 fps, showing that it is capable of real-time performance on an edge device. The contributions of this thesis are 1. Four low-memory event segmentation architectures. 2. A thorough analysis comparing the accuracies of EDGE-STREAMER to the state-of-the-art. 3. Experimental data showing the viability of EDGE-STREAMER on edge hardware. These results open the door for embedded device-based cameras, such as those used for surveillance, autonomous drones, and wearables, and provide possibilities for future work, including multi-modal streaming and embedded-device continual learning."]},{"key":"dc:title","label":"Title","values":["EDGE-STREAMER: A Lightweight, Self-Supervised Event Segmentation Model"]}]}],"canonical_facts":{"dc:contributor":["Sudeep Sarkar, Ph.D.","Shaun Canavan, Ph.D.","Mauricio Pamplona Segundo, Ph.D."],"dc:creator":["Miller, Lucas"],"dc:date":["2025-07-08T07:00:00Z"],"dc:description":["Event segmentation is the practice of autonomously detecting the boundaries of semantically connected sequences of actions within a video. It is connected to many areas of computer vision, including action recognition and event understanding. Nearly all current work requires the entire video to be stored in memory to be processed in multiple passes. This takes up valuable resources, increases processing time, prevents the use of low-power, low-memory devices, and precludes long-form content and live videos from undergoing event segmentation. This thesis introduces EDGE-STREAMER, a real-time, lightweight, self-supervised, transformer architecture capable of performing event segmentation in a single pass. EDGE-STREAMER uses techniques from reinforcement learning models to store representative information about the past to predict events, rather than the entire video. It takes advantage of weight sharing to reduce model size, further reducing its memory footprint and its training requirements. During training the model processes unlabeled video data in a single pass, producing event boundaries in real time. On Epic-Kitchens EDGE-STREAMER achieves the same MoF performance as STREAMER within 2\\%. On an NVIDIA Jetson Orin Nano, it sustains over 16 fps, showing that it is capable of real-time performance on an edge device. The contributions of this thesis are 1. Four low-memory event segmentation architectures. 2. A thorough analysis comparing the accuracies of EDGE-STREAMER to the state-of-the-art. 3. Experimental data showing the viability of EDGE-STREAMER on edge hardware. These results open the door for embedded device-based cameras, such as those used for surveillance, autonomous drones, and wearables, and provide possibilities for future work, including multi-modal streaming and embedded-device continual learning."],"dc:identifier":["https://digitalcommons.usf.edu/etd/10887"],"dc:subject":["Transformers","Online Machine Learning","Edge Computing","Event Understanding","Representation Learning","Computer Engineering"],"dc:title":["EDGE-STREAMER: A Lightweight, Self-Supervised Event Segmentation Model"],"thesis:degree_discipline":["Computer Science and Engineering"],"thesis:degree_level":["Thesis","M.S.Cp."],"thesis:degree_name":["MS in Computer Engineering (M.S.C.P.)"],"thesis:institution_name":["University of South Florida"]},"updated_at":"2026-11-19T08:49:48Z"}