Masked γ-SSL: learning uncertainty estimation via masked image modeling
This work proposes a semantic segmentation network that produces high-quality uncertainty estimates in a single forward pass. We exploit general representations from foundation models and unlabelled datasets through a Masked Image Modeling (MIM) approach, which is robust to augmentation hyper-parame...
| Päätekijät: | , , , |
|---|---|
| Aineistotyyppi: | Conference item |
| Kieli: | English |
| Julkaistu: |
IEEE
2024
|
| _version_ | 1826314918912065536 |
|---|---|
| author | Williams, DSW Gadd, M Newman, P De Martini, D |
| author_facet | Williams, DSW Gadd, M Newman, P De Martini, D |
| author_sort | Williams, DSW |
| collection | OXFORD |
| description | This work proposes a semantic segmentation network that produces high-quality uncertainty estimates in a single forward pass. We exploit general representations from foundation models and unlabelled datasets through a Masked Image Modeling (MIM) approach, which is robust to augmentation hyper-parameters and simpler than previous techniques. For neural networks used in safety-critical applications, bias in the training data can lead to errors; therefore it is crucial to understand a network’s limitations at run time and act accordingly. To this end, we test our proposed method on a number of test domains including the SAX Segmentation benchmark, which includes labelled test data from dense urban, rural and off-road driving domains. The proposed method consistently outperforms uncertainty estimation and Out-of-Distribution (OoD) techniques on this difficult benchmark. |
| first_indexed | 2024-03-07T08:29:09Z |
| format | Conference item |
| id | oxford-uuid:6a37d83e-421c-4b72-8da9-c5f75d115e04 |
| institution | University of Oxford |
| language | English |
| last_indexed | 2024-12-09T03:16:46Z |
| publishDate | 2024 |
| publisher | IEEE |
| record_format | dspace |
| spelling | oxford-uuid:6a37d83e-421c-4b72-8da9-c5f75d115e042024-10-21T11:48:42ZMasked γ-SSL: learning uncertainty estimation via masked image modelingConference itemhttp://purl.org/coar/resource_type/c_5794uuid:6a37d83e-421c-4b72-8da9-c5f75d115e04EnglishSymplectic ElementsIEEE2024Williams, DSWGadd, MNewman, PDe Martini, DThis work proposes a semantic segmentation network that produces high-quality uncertainty estimates in a single forward pass. We exploit general representations from foundation models and unlabelled datasets through a Masked Image Modeling (MIM) approach, which is robust to augmentation hyper-parameters and simpler than previous techniques. For neural networks used in safety-critical applications, bias in the training data can lead to errors; therefore it is crucial to understand a network’s limitations at run time and act accordingly. To this end, we test our proposed method on a number of test domains including the SAX Segmentation benchmark, which includes labelled test data from dense urban, rural and off-road driving domains. The proposed method consistently outperforms uncertainty estimation and Out-of-Distribution (OoD) techniques on this difficult benchmark. |
| spellingShingle | Williams, DSW Gadd, M Newman, P De Martini, D Masked γ-SSL: learning uncertainty estimation via masked image modeling |
| title | Masked γ-SSL: learning uncertainty estimation via masked image modeling |
| title_full | Masked γ-SSL: learning uncertainty estimation via masked image modeling |
| title_fullStr | Masked γ-SSL: learning uncertainty estimation via masked image modeling |
| title_full_unstemmed | Masked γ-SSL: learning uncertainty estimation via masked image modeling |
| title_short | Masked γ-SSL: learning uncertainty estimation via masked image modeling |
| title_sort | masked γ ssl learning uncertainty estimation via masked image modeling |
| work_keys_str_mv | AT williamsdsw maskedgssllearninguncertaintyestimationviamaskedimagemodeling AT gaddm maskedgssllearninguncertaintyestimationviamaskedimagemodeling AT newmanp maskedgssllearninguncertaintyestimationviamaskedimagemodeling AT demartinid maskedgssllearninguncertaintyestimationviamaskedimagemodeling |