Curriculum Learning (CL) is a training paradigm inspired by the structured learning progression in human curricula, where models are trained from easier to more difficult data [6]. It consists of two components: a Difficulty Measurer and a Training Scheduler [53], and is categorized into predefined CL [17, 21, 24, 32, 34, 40, 44, 50, 51] and automatic CL [20, 22, 25, 26, 37, 47, 48, 54, 57, 77ā79]. Predefined CL follows a manually designed training sequence, while automatic CL dynamically adjusts learning based on model feedback, offering greater flexibility [25]. Unlike existing automatic CL methods that focuse on full or other weakly supervision, this work is the first to explore its derivation to single point supervision, specifically for infrared small target detection. At the same time, conventional automatic CL suffers from early-stage misjudgments in sample difficulty [53]. The negative impact on our task will become more significant. Additionally, the single-point label format necessitates a rethinking of Difficulty Measurer and Training Scheduler design.