Reproduce “SatEdit: Mask-Conditioned Image Editing via VLM-Guided Segment Annotation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Muhammad Talha, Muhammad Ahmed Amer
SatEdit introduces a framework for mask-conditioned object-level satellite image editing by generating scalable training data from unlabeled imagery using a VLM-assisted segment annotation pipeline. This approach enabled the fine-tuning of an image editing model to achieve superior semantic alignment and spatial precision for object addition and removal tasks compared to baselines.
Key insights
- Leveraging existing foundation models (SAM2, VLM, Inpainting) in a structured pipeline effectively mitigates the data scarcity challenge for mask-conditioned satellite image editing.
- Constrained VLM labeling on mask-aware crops significantly reduces hallucinations and improves semantic consistency for object-level annotation.
- LoRA fine-tuning with an appropriate scale (e.g., ".6-.7") provides a balanced trade-off between semantic edit strength, background preservation, and leakage, avoiding over-adaptation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.