Lungu, Iustin (2025) 6D pose estimation of novel objects using AI generated meshes and FoundationPose. Bachelor's Thesis, Artificial Intelligence.
|
Text
IustinLunguBachelorThesis.pdf Download (35MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
toestemming lungu.pdf Restricted to Registered users only Download (227kB) |
Abstract
Robust 6D object pose estimation in real-world settings is often constrained by the availability of high-fidelity computer-aided design (CAD) models. This thesis investigates whether generative artificial intelligence (GenAI) models can bridge this gap by reconstructing viable 3D meshes from single-view RGB (red–green–blue) images. We integrate meshes generated by diffusion-based methods, Magic123, Zero123, and DreamFusion, into a pose estimation pipeline based on FoundationPose, and evaluate their effectiveness on 6D pose accuracy. Experiments are conducted on two fundamentally different datasets: the structured LINEMOD benchmark, which provides known objects with CAD models, and the unstructured HOTS dataset, which features novel, cluttered scenes without any available 3D geometry. Our results show that, despite variability in reconstruction quality, GenAI meshes can support reliable pose estimation under high-fidelity reconstructions, but coarse or noisy outputs lead to large failures, highlighting the need for improved reconstruction fidelity or more robust refinement. These findings support the use of generative models as viable substitutes for CAD assets, advancing scalable 6D pose estimation in unconstrained, real-world domains. The source code for this thesis is publicly available at our GitHub Repository: https://github.com/JustinLungu/FoundationPose-BachelorThesis/tree/main.
| Item Type: | Thesis (Bachelor's Thesis) |
|---|---|
| Supervisor name: | Mohades Kasaei, S.H. |
| Degree programme: | Artificial Intelligence |
| Thesis type: | Bachelor's Thesis |
| Language: | English |
| Date Deposited: | 09 Jul 2025 14:02 |
| Last Modified: | 09 Jul 2025 14:02 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/36043 |
Actions (login required)
 |
View Item |