filnow/

furniture-caption-qwen-lora

🤖 Qwen2-VL 2B Furniture Analysis LoRA

Model Description

🎯 Capabilities

• 🪑 Furniture Type Classification: Accurate identification of beds, tables, sofas, and chairs
• 🎨 Style Recognition: Design style categorization (modern, traditional, minimalist, etc.)
• 🌈 Color Analysis: Predominant color identification and description
• 🔨 Material Detection: Recognition of wood, metal, fabric, leather, and composite materials
• 📐 Shape Characterization: Physical form and geometric properties
• ✨ Detail Extraction: Decorative elements, hardware, and functional features
• 🏠 Room Context: Appropriate room placement recommendations
• 💰 Price Estimation: Relative price range categorization

⚙️ Technical Specifications

• Architecture: Vision-Language Transformer with LoRA adapters
• Input Resolution: 448x448 pixels (optimized for dataset)
• Output Format: Structured JSON with predefined attribute schema
• Memory Footprint: Significantly reduced compared to full fine-tuning
• Inference Speed: Optimized for real-time furniture analysis

📊 Training Details

• Training Images: 9,000 synthetic furniture images
• Validation: 1,000 real furniture photographs
• Image Generation: Stable Diffusion Medium 3.5
• Test Set Annotation: Qwen2 VL 72B
• Categories: Bed, table, sofa, chair

⚠️ Limitations

• Category Scope: Limited to four main furniture categories (bed, table, sofa, chair)
• Synthetic Training Bias: Potential domain gap between synthetic training data and real-world furniture
• Language Support: Optimized for English descriptions
• Image Quality: Best performance on well-lit, clear furniture images

📦 Model Artifacts

• LoRA Adapters: Lightweight adaptation weights for efficient deployment
• Configuration Files: Training hyperparameters and model settings
• Evaluation Metrics: Performance benchmarks on test dataset
• Example Outputs: Sample structured responses for reference