| Bibliography | Nadgouda, Chinmay Surendra: Open-Set 3D Scene Graph Generation with Fine-Grained Scene Understanding.
University of Stuttgart, Faculty of Computer Science, Electrical Engineering, and Information Technology, Master Thesis No. 27 (2025).
67 pages, english.
|
|---|
| Abstract | The abilityofarobottointeractwithitssurroundingsgreatlybenefitsfromasemanticallyrichscene graph.However,thisrequiresthescenegraphtohavesemanticinformationabouttheobjectsas wellastheirfunctionalinteractiveelements,alsoknownasparts.ConceptGraphs,astate-of-the-art piece ofresearch,fulfillsthefirstrequirementbutlacksthefine-grainedsegmentationofobject partsneededtofulfillthesecondrequirement.Inthisthesis,weintroduceanovelextensiontothe implementation ofConceptGraphs.WeproposetointegrateMask3DwithConceptGraphs.Wetrain the Mask3DmodelonadatasetresultingfromthemergerofARKitLabelMakerandSceneFun3D datasets. Thismodel,capableofpart-objectsegmentation,willenhanceConceptGraph’sabilityto capture fine-grainedinformationaboutobjectparts.Wealsotesthowwelloursystemworksby doing thetasksoffunctionalitysegmentationandtask-drivenaffordancegroundingthataredefined in SceneFun3D.Wepushforwardthestate-of-the-artperformanceontheSceneFun3Ddataset with ourimplementedsystem,demonstratingtheefficacyofourdevelopedsystem.Theresults obtained showanincreaseof6-9%inAveragePrecision(AP)atmeanIntersectionoverUnion (mIoU) thresholdsof25%and50%andover12%increaseinAP,theaverageoverdifferentIoU thresholds from0.5to0.95withastepof0.05.
|
|---|
| Department(s) | University of Stuttgart, Institute of Artificial Intelligence, Autonomous Systems
|
|---|
| Superviser(s) | Arras, Prof. Kai; Roitberg, Jun.-Prof. Alina; Rotondi, Dennis |
|---|
| Entry date | August 13, 2025 |
|---|