Multimodal Image Segmentation: An Analysis of a Modular Framework for Representation, Clustering, and Vision—Language Alignment
Fuente:
Zenodo
Saved in:
| Main Author: | |
|---|---|
| Format: | Recurso digital |
| Language: | English |
| Published: |
Zenodo
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866901879809638400 |
|---|---|
| author | Masson, Murillo |
| author_facet | Masson, Murillo |
| contents | <p>This work presents an exploratory analysis of a modular pipeline for multimodal image segmentation and labeling. The proposed framework integrates region proposals, deep visual representations extracted from DINOv2 and ResNet-50, unsupervised clustering, and vision-language alignment through CLIP.</p> <p>The study focuses on analyzing how different representation spaces and validation mechanisms influence the organization of candidate regions and the resulting segmentation outputs. The experiments examine multiple configurations of the pipeline, highlighting how semantic similarity and spatial consistency interact to shape the final selection of regions.</p> <p>The results show that different representation spaces induce distinct clustering patterns and spatial partitions, and that the combination of visual and semantic signals affects the consistency and diversity of the resulting segments.</p> <p>Code available <a target="_blank" rel="noopener">here</a>.</p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_19394225 |
| institution | Zenodo |
| language | eng |
| publishDate | 2026 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | Multimodal Image Segmentation: An Analysis of a Modular Framework for Representation, Clustering, and Vision—Language Alignment Masson, Murillo computer vision clustering DINOv2 SAM CLIP ResNet-50 <p>This work presents an exploratory analysis of a modular pipeline for multimodal image segmentation and labeling. The proposed framework integrates region proposals, deep visual representations extracted from DINOv2 and ResNet-50, unsupervised clustering, and vision-language alignment through CLIP.</p> <p>The study focuses on analyzing how different representation spaces and validation mechanisms influence the organization of candidate regions and the resulting segmentation outputs. The experiments examine multiple configurations of the pipeline, highlighting how semantic similarity and spatial consistency interact to shape the final selection of regions.</p> <p>The results show that different representation spaces induce distinct clustering patterns and spatial partitions, and that the combination of visual and semantic signals affects the consistency and diversity of the resulting segments.</p> <p>Code available <a target="_blank" rel="noopener">here</a>.</p> |
| title | Multimodal Image Segmentation: An Analysis of a Modular Framework for Representation, Clustering, and Vision—Language Alignment |
| topic | computer vision clustering DINOv2 SAM CLIP ResNet-50 |
| url | https://doi.org/10.5281/zenodo.19394225 |