Research map / Label-efficient and robust learning / Self-supervised and contrastive
Fine-Grained Segmentation Networks: Self-Supervised Segmentation for Improved Long-Term Visual Localization
Mans Larsson, Erik Stenborg, Carl Toft, Lars Hammarstrand, Torsten Sattler, Fredrik Kahl
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use FGSN with 100 clusters instead of 19 Cityscapes classes | SSMC · Extended CMU Seasons, Urban | localization % at 0.25 m / 2 deg: 71.8 -> 81.8 | Improved |
| use FGSN with 100 clusters instead of 19 Cityscapes classes | SSMC · Extended CMU Seasons, Park | localization % at 0.25 m / 2 deg: 32.8 -> 51.3 | Improved |
| use FGSN with 200 clusters instead of 19 Cityscapes classes | SSMC · RobotCar Seasons, night | localization % at 0.25 m / 2 deg: 8.2 -> 9.6 | Improved |
| increase clusters from 100 to 1000 (ablation) | SSMC with FGSN · Extended CMU Seasons, Urban | localization % at 0.25 m / 2 deg: 81.8 -> 78.0 | Hurt |
| increase clusters from 100 to 1000 (ablation) | SSMC with FGSN · Extended CMU Seasons, Park | localization % at 0.25 m / 2 deg: 51.3 -> 45.1 | Hurt |
| initialize with segmentation pretraining instead of classification pretraining (200 clusters) (ablation) | FGSN with SSMC · Extended CMU Seasons, Urban | localization % at 0.25 m / 2 deg: 70.8 -> 81.0 | Improved |
| initialize with segmentation pretraining instead of classification pretraining (1000 clusters) (ablation) | FGSN with SSMC · Extended CMU Seasons, Urban | localization % at 0.25 m / 2 deg: 47.4 -> 78.0 | Improved |
| add correspondence loss (200 clusters) (ablation) | FGSN with SSMC · Extended CMU Seasons, Urban | localization % at 0.25 m / 2 deg: 78.7 -> 81.0 | Improved |
| add correspondence loss (1000 clusters) (ablation) | FGSN with SSMC · Extended CMU Seasons, Urban | localization % at 0.25 m / 2 deg: 73.4 -> 78.0 | Improved |
| add correspondence loss to 66-class Vistas segmentation network (ablation) | SSMC with semantic classes · Extended CMU Seasons, Urban | localization % at 0.25 m / 2 deg: 65.8 -> 66.5 | No clear effect |
| do not repeat/reset clustering during training (100 clusters) (ablation) | FGSN with SSMC · Extended CMU Seasons, Urban | localization % at 0.25 m / 2 deg: 81.8 -> 85.3 | Improved |
| use FGSN 200 clusters in place of semantic segmentation | PFSL · Extended CMU Seasons, Park | localization % at 0.25 m / 2 deg: 39.0 -> 64.8 | Improved |
Closest papers by meaning
- VS-Net: Voting With Segmentation for Visual Localization (CVPR 2021 · similarity 0.87)
- Interpretable and Accurate Fine-grained Recognition via Region Grouping (CVPR 2020 · similarity 0.87)
- Self-supervising Fine-grained Region Similarities for Large-scale Image Localization (ECCV 2020 · similarity 0.87)
- Dynamic Position-aware Network for Fine-grained Image Recognition (AAAI 2021 · similarity 0.86)
- Object Instance Segmentation and Fine-Grained Localization Using Hypercolumns (TPAMI 2016 · similarity 0.86)
- Revisiting Dilated Convolution: A Simple Approach for Weakly- and Semi-Supervised Semantic Segmentation (CVPR 2018 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
