Insights from Autoresearch for Solar Panel Segmentation

Justinas Lekavicius, Kursat Komurcu, Valentas Gruzauskas, Linas Petkevicius

YAYIN:7 Eki 2026 ALAN:cs.CV OKUMA:24

Özet

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.

Özetle: 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.

Özet

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.

Orijinal Özet (İngilizce)

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.

Kaynak: arXiv:2610.10491 · PDF

BibTeX

@article{lekavicius2026insights,
  title   = {Insights from Autoresearch for Solar Panel Segmentation},
  author  = {Justinas Lekavicius and Kursat Komurcu and Valentas Gruzauskas and Linas Petkevicius},
  journal = {arXiv preprint arXiv:2610.10491},
  year    = {2026},
  url     = {https://arxiv.org/abs/2610.10491}
}

Tartışma

Bu habere emoji ile tepki ver

İptal

Hizli:

Henüz yorum yok. İlk yorumu siz yapın!

Yapıcı ve saygılı yorumlar bekliyoruz. Topluluk kuralları