Groefsema, Steff (2023) Uncertainty Quantification in DETR for Pedestrian Detection. Master's Thesis / Essay, Artificial Intelligence.
|
Text
mAI_2023_GroefsemaS.pdf Download (13MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
toestemming.pdf Restricted to Registered users only Download (132kB) |
Abstract
In this research we investigate Uncertaintly Learning within DETR and a Single-Shot multibox Detector(SSD) on four different data sets. Because Pascal VOC and MS COCO have different images in comparison with Crowd Human and Wider Person, we hope to find different results in terms of mean average precision and bounding box deviation. The research questions are: Does applying a Deep sub-ensemble structure on the existing DETR architecture make the predictions of the total network more robust on unseen data? How well does DETR generalize with a deep sub-ensemble? If we train DETR on data set A, how well will it perform on other unseen data sets? Does applying a Deep sub-ensemble structure on the existing SSD architecture make the predictions of the total network more robust on unseen data? How well does SSD generalize with a deep sub-ensemble? If we train SSD on data set A, how well will it perform on other unseen data sets? We hypothesize a sub-ensemble will make DETR more robust. We think sub-ensembles will improve the generalization ability of DETR. We think implementing a sub-ensemble with SSD will make the network more robust. Lastly we expect the SSD will suffer less from overfitting the training data in comparison to DETR. We found no evidence the deep sub-ensembles improved the accuracy scores, nor the generaliza- tion ability of DETR and SSD. We discuss the results obtained and discuss points for future research.
| Item Type: | Thesis (Master's Thesis / Essay) |
|---|---|
| Supervisor name: | Valdenegro Toro, M.A. and Mohades Kasaei, S.H. |
| Degree programme: | Artificial Intelligence |
| Thesis type: | Master's Thesis / Essay |
| Language: | English |
| Date Deposited: | 28 Mar 2023 14:07 |
| Last Modified: | 28 Mar 2023 14:07 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/29479 |
Actions (login required)
 |
View Item |