Multi-Point Positional Insertion Tuning for Small Object Detection

TL;DR

This paper introduces multi-point positional insertion (MPI) tuning, a parameter-efficient method that enhances small object detection by integrating multiple positional embeddings into frozen pretrained models, reducing tuning parameters while maintaining performance.

Contribution

The paper proposes MPI tuning, a novel PEFT approach that efficiently improves small object detection by adding multiple positional embeddings to pretrained models.

Findings

MPI achieves comparable performance to existing PEFT methods.

MPI significantly reduces the number of tunable parameters.

Effective on the SODA-D dataset.

Abstract

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.