GFT: Graph Feature Tuning for Efficient Point Cloud Analysis

TL;DR

This paper introduces Graph Features Tuning (GFT), a parameter-efficient method for point cloud analysis that learns dynamic graph features to reduce trainable parameters while maintaining high performance.

Contribution

GFT is a novel point-cloud-specific PEFT approach that learns dynamic graphs with lightweight GCNs and integrates them into transformer models for efficient analysis.

Findings

GFT achieves comparable performance to existing methods.

GFT significantly reduces the number of trainable parameters.

GFT is effective for object classification and segmentation.

Abstract

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.