The First Competition on Resource-Limited Infrared Small Target Detection Challenge: Methods and Results

TL;DR

This paper documents the inaugural LimitIRSTD competition for resource-limited infrared small target detection, with Track 1 focusing on weak supervision under a single coarse point and Track 2 on lightweight, pixel-level supervision. It introduces the WideIRSTD dataset family and a dual-track evaluation framework that blends traditional detection metrics with resource constraints, guiding fair comparisons. Key contributions include a survey of two main methodological themes—label generation/refinement (LESPS-based and pseudo-label approaches) and lightweight network design (distillation, pruning, and efficient architectures)—as well as detailed results from top teams across both tracks. The findings demonstrate substantial progress toward robust IRSTD under limited resources and underscore the value of high-quality labeling, data filtering, and efficient architectures for practical deployment.

Abstract

In this paper, we briefly summarize the first competition on resource-limited infrared small target detection (namely, LimitIRSTD). This competition has two tracks, including weakly-supervised infrared small target detection (Track 1) and lightweight infrared small target detection (Track 2). 46 and 60 teams successfully registered and took part in Tracks 1 and Track 2, respectively. The top-performing methods and their results in each track are described with details. This competition inspires the community to explore the tough problems in the application of infrared small target detection, and ultimately promote the deployment of this technology under limited resource.

Figures (16)

• Figure 1: An illustration of WideIRSTD dataset.
• Figure 2: An illustration of WideIRSTD dataset.
• Figure 3: An illustration of the overall framework of the Chainey Team.
• Figure 4: An illustration of the overall framework of the XJTU-IR Team.
• Figure 5: An illustration of the overall framework of the MCV-TEAM.