mohan2/StyleTransfer-SFT-OmniStyle

--- license: apache-2.0 task_categories: - image-to-image tags: - style-transfer - supervised-fine-tuning - omnistyle size_categories: - 100K<n<1M --- # StyleTransfer-SFT-OmniStyle **OmniStyle-150K** triplet dataset for style transfer SFT training. ## 📊 Dataset Statistics - **Stylized results**: 143,992 images - **Content images**: 1,812 - **Style images**: 950 ## 📦 Structure ``` OmniStyle-150k/ ├── content/ # Original content images ├── style/ # Style reference images └── OmniStyle-150K/ # Stylized results └── <content>&&<style>.png ``` ## 🚀 Usage ```bash # Merge and extract cat omnistyle_150k.tar.part_* > omnistyle_150k.tar tar -xvf omnistyle_150k.tar ``` ```python import os for img in os.listdir("OmniStyle-150k/OmniStyle-150K"): content_name, style_name = img.split("&&") style_name = style_name[:-4] # content: OmniStyle-150k/content/{content_name} # style: OmniStyle-150k/style/{style_name} # result: OmniStyle-150k/OmniStyle-150K/{img} ``` ## 📝 Triplet Format Each stylized image filename: `<content_name>&&<style_name>.png` - **Content**: Input image to be stylized - **Style**: Reference style image - **Result**: Stylized output