{"id":{"repo_id":"csusb","oai_identifier":"oai:scholarworks.lib.csusb.edu:etd-3644"},"canonical_url":"https://search.ndltd.org/etd/csusb/oai:scholarworks.lib.csusb.edu:etd-3644","repository":{"repo_id":"csusb","name":"CSUniversity San Bernardino","base_url":"https://scholarworks.lib.csusb.edu/do/oai/"},"display":{"title":"ROBUST MULTI-OBJECT DETECTION UNDER LOW-LIGHT CONDITIONS: A COMPARATIVE STUDY OF TRANSFORMER-BASED DETECTORS WITH LOW-LIGHT AUGMENTED FINE-TUNING","abstract":"Object detection in low-light environments remains a significant challenge for modern computer vision systems. This paper presents a systematic comparative study of three state-of-the-art transformer-based object detection architectures — DETR (DEtection TRansformer), YOLOS (You Only Look at One Sequence), and RT-DETR (Real-Time DEtection TRansformer) — evaluated under both normal and degraded lighting conditions. We propose a three-phase experimental methodology: (1) pretrained baseline evaluation, (2) fine-tuning with low-light augmented training data, and (3) dual test set evaluation under normal and simulated low-light conditions. Our experiments on the COCO val2017 dataset demonstrate that low-light augmented fine-tuning yields substantial improvements across all three architectures, with YOLOS-Small achieving the largest relative improvement (+991% mAP increase) and RT-DETR-R50 maintaining the highest absolute performance (mAP = 0.1232 on normal, 0.1227 on low-light). Notably, all models exhibit strong robustness to low-light degradation after augmented training, with minimal performance gaps between normal and low-light evaluation conditions. These findings suggest that data augmentation strategies incorporating synthetic low-light simulation are an effective and architecture-agnostic approach to improving detection robustness in challenging illumination scenarios.","abstract_html":"Object detection in low-light environments remains a significant challenge for modern computer vision systems. This paper presents a systematic comparative study of three state-of-the-art transformer-based object detection architectures — DETR (DEtection TRansformer), YOLOS (You Only Look at One Sequence), and RT-DETR (Real-Time DEtection TRansformer) — evaluated under both normal and degraded lighting conditions. We propose a three-phase experimental methodology: (1) pretrained baseline evaluation, (2) fine-tuning with low-light augmented training data, and (3) dual test set evaluation under normal and simulated low-light conditions. Our experiments on the COCO val2017 dataset demonstrate that low-light augmented fine-tuning yields substantial improvements across all three architectures, with YOLOS-Small achieving the largest relative improvement (+991% mAP increase) and RT-DETR-R50 maintaining the highest absolute performance (mAP = 0.1232 on normal, 0.1227 on low-light). Notably, all models exhibit strong robustness to low-light degradation after augmented training, with minimal performance gaps between normal and low-light evaluation conditions. These findings suggest that data augmentation strategies incorporating synthetic low-light simulation are an effective and architecture-agnostic approach to improving detection robustness in challenging illumination scenarios.","abstract_has_math":false,"creators":["Zhang, Naiqi"],"institution":null,"degree_name":"Master of Science in Computer Science","degree_level":"Restricted Project: Campus only access","degree_discipline":"School of Computer Science and Engineering","degree_department":null,"school":null,"contributors":["Yan Zhang"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":null,"date_issued":"","date_published":null,"updated_at":"2026-11-19T08:10:18Z","subjects":["Computer Vision","Object Detection","Fine Tuning","Transformer","Low-Light Augmentation","Other Computer Engineering","Signal Processing"],"languages":[],"rights":[],"rights_urls":[],"identifier_entries":[],"person_identifier_entries":[]},"links":{"outbound_url":"https://scholarworks.lib.csusb.edu/etd/2415","outbound_label":"Repository record","outbound_source":"dc:identifier","record_url":"https://scholarworks.lib.csusb.edu/etd/2415","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://scholarworks.lib.csusb.edu/do/oai/?verb=GetRecord&metadataPrefix=oai_etdms&identifier=oai%3Ascholarworks.lib.csusb.edu%3Aetd-3644","prefix":"oai_etdms"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Yan Zhang"]},{"key":"dc:creator","label":"Author","values":["Zhang, Naiqi"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.available","label":"Dc Date Available","values":["2026-04-08T07:00:00Z"]},{"key":"dc:date.created","label":"Dc Date Created","values":["2026-05-01T07:00:00Z"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["School of Computer Science and Engineering"]},{"key":"thesis:degree_level","label":"Degree Level","values":["Restricted Project: Campus only access"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Master of Science in Computer Science"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Computer Vision","Object Detection","Fine Tuning","Transformer","Low-Light Augmentation","Other Computer Engineering","Signal Processing"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["https://scholarworks.lib.csusb.edu/etd/2415"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Object detection in low-light environments remains a significant challenge for modern computer vision systems. This paper presents a systematic comparative study of three state-of-the-art transformer-based object detection architectures — DETR (DEtection TRansformer), YOLOS (You Only Look at One Sequence), and RT-DETR (Real-Time DEtection TRansformer) — evaluated under both normal and degraded lighting conditions. We propose a three-phase experimental methodology: (1) pretrained baseline evaluation, (2) fine-tuning with low-light augmented training data, and (3) dual test set evaluation under normal and simulated low-light conditions. Our experiments on the COCO val2017 dataset demonstrate that low-light augmented fine-tuning yields substantial improvements across all three architectures, with YOLOS-Small achieving the largest relative improvement (+991% mAP increase) and RT-DETR-R50 maintaining the highest absolute performance (mAP = 0.1232 on normal, 0.1227 on low-light). Notably, all models exhibit strong robustness to low-light degradation after augmented training, with minimal performance gaps between normal and low-light evaluation conditions. These findings suggest that data augmentation strategies incorporating synthetic low-light simulation are an effective and architecture-agnostic approach to improving detection robustness in challenging illumination scenarios."]},{"key":"dc:title","label":"Title","values":["ROBUST MULTI-OBJECT DETECTION UNDER LOW-LIGHT CONDITIONS: A COMPARATIVE STUDY OF TRANSFORMER-BASED DETECTORS WITH LOW-LIGHT AUGMENTED FINE-TUNING"]}]}],"canonical_facts":{"dc:contributor":["Yan Zhang"],"dc:creator":["Zhang, Naiqi"],"dc:date.available":["2026-04-08T07:00:00Z"],"dc:date.created":["2026-05-01T07:00:00Z"],"dc:description.abstract":["Object detection in low-light environments remains a significant challenge for modern computer vision systems. This paper presents a systematic comparative study of three state-of-the-art transformer-based object detection architectures — DETR (DEtection TRansformer), YOLOS (You Only Look at One Sequence), and RT-DETR (Real-Time DEtection TRansformer) — evaluated under both normal and degraded lighting conditions. We propose a three-phase experimental methodology: (1) pretrained baseline evaluation, (2) fine-tuning with low-light augmented training data, and (3) dual test set evaluation under normal and simulated low-light conditions. Our experiments on the COCO val2017 dataset demonstrate that low-light augmented fine-tuning yields substantial improvements across all three architectures, with YOLOS-Small achieving the largest relative improvement (+991% mAP increase) and RT-DETR-R50 maintaining the highest absolute performance (mAP = 0.1232 on normal, 0.1227 on low-light). Notably, all models exhibit strong robustness to low-light degradation after augmented training, with minimal performance gaps between normal and low-light evaluation conditions. These findings suggest that data augmentation strategies incorporating synthetic low-light simulation are an effective and architecture-agnostic approach to improving detection robustness in challenging illumination scenarios."],"dc:identifier":["https://scholarworks.lib.csusb.edu/etd/2415"],"dc:subject":["Computer Vision","Object Detection","Fine Tuning","Transformer","Low-Light Augmentation","Other Computer Engineering","Signal Processing"],"dc:title":["ROBUST MULTI-OBJECT DETECTION UNDER LOW-LIGHT CONDITIONS: A COMPARATIVE STUDY OF TRANSFORMER-BASED DETECTORS WITH LOW-LIGHT AUGMENTED FINE-TUNING"],"thesis:degree_discipline":["School of Computer Science and Engineering"],"thesis:degree_level":["Restricted Project: Campus only access"],"thesis:degree_name":["Master of Science in Computer Science"]},"updated_at":"2026-11-19T08:10:18Z"}