GFT: Graph Feature Tuning for Efficient Point Cloud Analysis
This work addresses the need for more efficient adaptation of models to point cloud tasks, offering a domain-specific improvement over existing methods.
The paper tackles the problem of reducing trainable parameters for efficient point cloud analysis by proposing Graph Features Tuning (GFT), a point-cloud-specific parameter-efficient fine-tuning method that learns dynamic graphs and uses skip connections and cross-attention, achieving competitive performance on object classification and segmentation tasks.
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 at https://github.com/manishdhakal/GFT.