aimhkimi74/Bert-Model-living-room

🏡 Fine-Tuned BERT for Interior Design (Living Room)

This is a fine-tuned BERT model for interior design prompt validation and style classification. It was developed as part of my Final Year Project: Text-to-Image Interior Design Generator with Generative AI Assistance.

The model supports:

• —✅ Prompt Validation – distinguish valid vs. invalid design prompts
• —🏷️ Style Classification – classify valid prompts into 7 living room design styles:

• —Modern
• —Scandinavian
• —Rustic
• —Industrial
• —Traditional
• —Mid-Century Modern
• —Coastal

📊 Results

• —Prompt Validation (binary classification) → F1-score: 1.00
• —Style Classification (7 classes) → F1-score: 0.99

📂 Model Details

• —Base model: bert-base-uncased
• —Library: Transformers (PyTorch)
• —Trained with: custom dataset of interior design prompts
• —Labels mapping:

  {
    "id2label": {
      "0": "Modern",
      "1": "Scandinavian",
      "2": "Rustic",
      "3": "Industrial",
      "4": "Traditional",
      "5": "Mid-Century Modern",
      "6": "Coastal"
    },
    "label2id": {
      "Modern": 0,
      "Scandinavian": 1,
      "Rustic": 2,
      "Industrial": 3,
      "Traditional": 4,
      "Mid-Century Modern": 5,
      "Coastal": 6
    }
  }

⚙️ How to Use

from transformers import BertTokenizer, BertForSequenceClassification
import torch

# Load model from Hugging Face Hub
tokenizer = BertTokenizer.from_pretrained("aimhkimi74/Bert-Model-living-room")
model = BertForSequenceClassification.from_pretrained("aimhkimi74/Bert-Model-living-room")

# Example prompt
text = "A modern living room with a gray sofa and wooden floor."
inputs = tokenizer(text, return_tensors="pt")
outputs = model(**inputs)
pred = torch.argmax(outputs.logits, dim=1)

# Map prediction to style
label_map = {
    0: "Modern",
    1: "Scandinavian",
    2: "Rustic",
    3: "Industrial",
    4: "Traditional",
    5: "Mid-Century Modern",
    6: "Coastal"
}
print("Predicted Style:", label_map[pred.item()])

📖 Training Procedure

• —Optimizer: AdamW
• —Learning Rate: 5e-5
• —Epochs: 4
• —Batch size: 32
• —Evaluation metrics: Accuracy, F1-score

⚠️ Intended Use & Limitations

• —Intended use: Assist a text-to-image system by validating prompts and tagging living-room styles.
• —Not intended for: Architectural safety decisions, non-living-room styles, or multilingual inputs (English only).
• —Known limits: Performance may drop outside the 7 styles or with very short/ambiguous prompts.

📜 License

This model is released under the MIT License.

📚 References

• —Devlin, J. et al. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding
• —Hugging Face Transformers library
• —Rombach, R. et al. High-Resolution Image Synthesis with Latent Diffusion Models