GFT: Graph Feature Tuning for Efficient Point Cloud Analysis

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Dhakal, Manish, Dasari, Venkat R., Sunderraman, Rajshekhar, Ding, Yi
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866915646271389696
author Dhakal, Manish
Dasari, Venkat R.
Sunderraman, Rajshekhar
Ding, Yi
author_facet Dhakal, Manish
Dasari, Venkat R.
Sunderraman, Rajshekhar
Ding, Yi
contents Parameter-efficient fine-tuning (PEFT) significantly reduces computational and memory costs by updating only a small subset of the model's parameters, enabling faster adaptation to new tasks with minimal loss in performance. Previous studies have introduced PEFTs tailored for point cloud data, as general approaches are suboptimal. To further reduce the number of trainable parameters, we propose a point-cloud-specific PEFT, termed Graph Features Tuning (GFT), which learns a dynamic graph from initial tokenized inputs of the transformer using a lightweight graph convolution network and passes these graph features to deeper layers via skip connections and efficient cross-attention modules. Extensive experiments on object classification and segmentation tasks show that GFT operates in the same domain, rivalling existing methods, while reducing the trainable parameters. Code is available at https://github.com/manishdhakal/GFT.
format Preprint
id arxiv_https___arxiv_org_abs_2511_10799
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle GFT: Graph Feature Tuning for Efficient Point Cloud Analysis
Dhakal, Manish
Dasari, Venkat R.
Sunderraman, Rajshekhar
Ding, Yi
Computer Vision and Pattern Recognition
Parameter-efficient fine-tuning (PEFT) significantly reduces computational and memory costs by updating only a small subset of the model's parameters, enabling faster adaptation to new tasks with minimal loss in performance. Previous studies have introduced PEFTs tailored for point cloud data, as general approaches are suboptimal. To further reduce the number of trainable parameters, we propose a point-cloud-specific PEFT, termed Graph Features Tuning (GFT), which learns a dynamic graph from initial tokenized inputs of the transformer using a lightweight graph convolution network and passes these graph features to deeper layers via skip connections and efficient cross-attention modules. Extensive experiments on object classification and segmentation tasks show that GFT operates in the same domain, rivalling existing methods, while reducing the trainable parameters. Code is available at https://github.com/manishdhakal/GFT.
title GFT: Graph Feature Tuning for Efficient Point Cloud Analysis
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2511.10799