Efficient Segment Anything with Depth-Aware Fusion and Limited Training Data
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2026
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866918333731831808 |
|---|---|
| author | Zhou, Yiming Xie, Xuenjie Li, Panfeng Kunz, Albrecht Osman, Ahmad Maldague, Xavier |
| author_facet | Zhou, Yiming Xie, Xuenjie Li, Panfeng Kunz, Albrecht Osman, Ahmad Maldague, Xavier |
| contents | Segment Anything Models (SAM) achieve impressive universal segmentation performance but require massive datasets (e.g., 11M images) and rely solely on RGB inputs. Recent efficient variants reduce computation but still depend on large-scale training. We propose a lightweight RGB-D fusion framework that augments EfficientViT-SAM with monocular depth priors. Depth maps are generated with a pretrained estimator and fused mid-level with RGB features through a dedicated depth encoder. Trained on only 11.2k samples (less than 0.1\% of SA-1B), our method achieves higher accuracy than EfficientViT-SAM, showing that depth cues provide strong geometric priors for segmentation. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_11804 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Efficient Segment Anything with Depth-Aware Fusion and Limited Training Data Zhou, Yiming Xie, Xuenjie Li, Panfeng Kunz, Albrecht Osman, Ahmad Maldague, Xavier Computer Vision and Pattern Recognition Image and Video Processing Segment Anything Models (SAM) achieve impressive universal segmentation performance but require massive datasets (e.g., 11M images) and rely solely on RGB inputs. Recent efficient variants reduce computation but still depend on large-scale training. We propose a lightweight RGB-D fusion framework that augments EfficientViT-SAM with monocular depth priors. Depth maps are generated with a pretrained estimator and fused mid-level with RGB features through a dedicated depth encoder. Trained on only 11.2k samples (less than 0.1\% of SA-1B), our method achieves higher accuracy than EfficientViT-SAM, showing that depth cues provide strong geometric priors for segmentation. |
| title | Efficient Segment Anything with Depth-Aware Fusion and Limited Training Data |
| topic | Computer Vision and Pattern Recognition Image and Video Processing |
| url | https://arxiv.org/abs/2602.11804 |