Multi-Point Positional Insertion Tuning for Small Object Detection
This work addresses the problem of efficient finetuning for small object detection in computer vision, offering an incremental improvement over existing parameter-efficient methods.
The paper tackled the computational and memory expense of finetuning large pretrained models for small object detection by introducing multi-point positional insertion (MPI) tuning, a parameter-efficient method that performed comparably to existing PEFT methods while significantly reducing tuned parameters.
Small object detection aims to localize and classify small objects within images. With recent advances in large-scale vision-language pretraining, finetuning pretrained object detection models has emerged as a promising approach. However, finetuning large models is computationally and memory expensive. To address this issue, this paper introduces multi-point positional insertion (MPI) tuning, a parameter-efficient finetuning (PEFT) method for small object detection. Specifically, MPI incorporates multiple positional embeddings into a frozen pretrained model, enabling the efficient detection of small objects by providing precise positional information to latent features. Through experiments, we demonstrated the effectiveness of the proposed method on the SODA-D dataset. MPI performed comparably to conventional PEFT methods, including CoOp and VPT, while significantly reducing the number of parameters that need to be tuned.