Introspection in Learned Semantic Scene Graph Localisation
Fuente:
arXiv
Salvato in:
| Autori principali: | , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866912636055060480 |
|---|---|
| author | Bissessur, Manshika Charvi Panagiotaki, Efimia De Martini, Daniele |
| author_facet | Bissessur, Manshika Charvi Panagiotaki, Efimia De Martini, Daniele |
| contents | This work investigates how semantics influence localisation performance and robustness in a learned self-supervised, contrastive semantic localisation framework. After training a localisation network on both original and perturbed maps, we conduct a thorough post-hoc introspection analysis to probe whether the model filters environmental noise and prioritises distinctive landmarks over routine clutter. We validate various interpretability methods and present a comparative reliability analysis. Integrated gradients and Attention Weights consistently emerge as the most reliable probes of learned behaviour. A semantic class ablation further reveals an implicit weighting in which frequent objects are often down-weighted. Overall, the results indicate that the model learns noise-robust, semantically salient relations about place definition, thereby enabling explainable registration under challenging visual and structural variations. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2510_07053 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Introspection in Learned Semantic Scene Graph Localisation Bissessur, Manshika Charvi Panagiotaki, Efimia De Martini, Daniele Machine Learning Artificial Intelligence Computer Vision and Pattern Recognition Robotics I.2.10; I.2.9; I.4.8; I.5.2; I.5.1 This work investigates how semantics influence localisation performance and robustness in a learned self-supervised, contrastive semantic localisation framework. After training a localisation network on both original and perturbed maps, we conduct a thorough post-hoc introspection analysis to probe whether the model filters environmental noise and prioritises distinctive landmarks over routine clutter. We validate various interpretability methods and present a comparative reliability analysis. Integrated gradients and Attention Weights consistently emerge as the most reliable probes of learned behaviour. A semantic class ablation further reveals an implicit weighting in which frequent objects are often down-weighted. Overall, the results indicate that the model learns noise-robust, semantically salient relations about place definition, thereby enabling explainable registration under challenging visual and structural variations. |
| title | Introspection in Learned Semantic Scene Graph Localisation |
| topic | Machine Learning Artificial Intelligence Computer Vision and Pattern Recognition Robotics I.2.10; I.2.9; I.4.8; I.5.2; I.5.1 |
| url | https://arxiv.org/abs/2510.07053 |