Layer-wise model merging for unsupervised domain adaptation in segmentation tasks

Merging parameters of multiple models has resurfaced as an effective strategy to enhance task performance and robustness, but prior work is limited by the high costs of ensemble creation and inference. In this paper, we leverage the abundance of freely accessible trained models to introduce a cost-f...

Descripción completa

Detalles Bibliográficos
Autores: Alcover Couso, Roberto, San Miguel Avedillo, Juan Carlos, Escudero Viñolo, Marcos, Martínez Sánchez, José María
Tipo de recurso: artículo
Fecha de publicación:2025
País:España
Institución:Universidad Autónoma de Madrid
Repositorio:Biblos-e Archivo. Repositorio Institucional de la UAM
Idioma:inglés
OAI Identifier:oai:repositorio.uam.es:10486/718572
Acceso en línea:http://hdl.handle.net/10486/718572
https://dx.doi.org/10.1007/s00371-025-03843-7
Access Level:acceso abierto
Palabra clave:Model Merging
Unsupervised Domain Adaptation
Semantic Segmentation
Panoptic Segmentation
Synthetic Data
Informática
Descripción
Sumario:Merging parameters of multiple models has resurfaced as an effective strategy to enhance task performance and robustness, but prior work is limited by the high costs of ensemble creation and inference. In this paper, we leverage the abundance of freely accessible trained models to introduce a cost-free approach to model merging. It focuses on a layer-wise integration of merged models, aiming to maintain the distinctiveness of the task-specific final layers while unifying the initial layers, which are primarily associated with feature extraction. This approach ensures parameter consistency across all layers, essential for boosting performance. Moreover, it facilitates seamless integration of knowledge, enabling effective merging of models from different datasets and tasks. Specifically, we investigate its applicability in unsupervised domain adaptation (UDA), an unexplored area for model merging, for semantic and panoptic segmentation. Experimental results demonstrate substantial UDA improvements without additional costs for merging same-architecture models from distinct datasets (↑ 2.6% mIoU) and different-architecture models with a shared backbone (↑ 6.8% mIoU). Furthermore, merging semantic and panoptic segmen tation models increases mPQ by 7%. These findings are validated across a wide variety of UDA strategies, architectures and datasets. The code will be publicly available upon acceptance in the LWMM repository: http://www-vpu.eps.uam.es/LWMM/