自动研究在太阳能电池板分割中的见解
Insights from Autoresearch for Solar Panel Segmentation
浏览论文内容
中文总结 AI 辅助
本文探讨AutoResearch协议在太阳能电池板分割中的应用,通过编码语言模型在GPU预算内迭代改进训练程序,发现其能提升基线性能但修改不跨硬件迁移,其中Qwen3-8B在真实图像上达到0.836的IoU,略优于GAN增强方案。
中文摘要 AI 辅助
本文研究了AutoResearch协议,在该协议中,编码语言模型在一小时的GPU预算内编辑训练程序,并且仅当验证IoU(交并比)提高时才保留更改。该协议应用于冻结的真实图像分割上的光伏电池板分割,其中DeepLabV3--ResNet-50保持固定。使用Gemma 4 12B和Qwen3-8B进行的三个24次实验活动均改善了其一小时的基线,但保留的修改无法跨硬件迁移。仅在真实图像上训练的Qwen3-8B配置达到了0.836的测试IoU,而参考的GAN增强方案为0.833。研究仓库见此https URL。
英文摘要
This paper investigates AutoResearch, a protocol in which a coding language model edits a training program under a one-hour GPU budget and retains a change only if validation IoU improves. The protocol is applied to photovoltaic panel segmentation on a frozen real-image split, with DeepLabV3--ResNet-50 held fixed. Three campaigns of 24 experiments, using Gemma~4 12B, Qwen3-8B all improve their one-hour baselines, but retained modifications do not transfer across hardware. The Qwen3-8B configuration, trained on real images only, reaches a test IoU of 0.836 versus 0.833 for the reference GAN-augmented schedule. Research repository https://github.com/VU-AIML/automl4eo-autoresearch-segmentation.
发表机构
- Vilnius University(维尔纽斯大学)
- IRISA, Université Bretagne Sud(IRISA,南布列塔尼大学)
- European Commission Joint Research Centre(欧盟委员会联合研究中心)
机构由 AI 辅助整理,请以论文原文为准。