| --- |
| license: mit |
| library_name: onnx |
| pipeline_tag: image-segmentation |
| tags: |
| - computer-vision |
| - onnx |
| - carpet-analysis |
| - defect-detection |
| - unet |
| - sam2 |
| - image-classification |
| --- |
| |
| # 🧶 Carpet Analysis System - Deep Learning Models |
|
|
| A collection of pre-trained deep learning models optimized for real-time carpet quality inspection, style classification, material detection, and surface segmentation. |
|
|
| Optimized with **ONNX Runtime** for low latency and efficient deployment on both CPU and GPU backends. |
|
|
| --- |
|
|
| ## 🚀 Live Demo & API Endpoint |
|
|
| These models power the active production backend deployed on Hugging Face Spaces: |
| * **Live Space:** [yosedie/carpet-analysis-api](https://huggingface.co/spaces/yosedie/carpet-analysis-api) |
| * **Interactive Swagger UI:** [https://yosedie-carpet-analysis-api.hf.space/docs](https://yosedie-carpet-analysis-api.hf.space/docs) |
|
|
| --- |
|
|
| ## 📦 Model Portfolio |
|
|
| | Model File | Architecture | Task / Objective | Input Resolution | Format | |
| | :--- | :--- | :--- | :--- | :--- | |
| | **`unet_model.onnx`** | U-Net | Carpet surface segmentation & defect coverage ratio calculation | 512x512 | ONNX | |
| | **`cnn_a_model.onnx`** | Custom CNN | Carpet Style Classification (Asia, Malaysia, Mie, Persia, Polos, Sejadah) | 224x224 | ONNX | |
| | **`cnn_b_model.onnx`** | Custom CNN | Fiber Material Identification (Polypropylene, Nylon, Polyester) | 224x224 | ONNX | |
| | **`sam2_hiera_base_plus.pt`** | SAM 2 (Hiera Base+) | Zero-shot promptable segmentation for fine-grained boundary extraction | Multi-scale | PyTorch / SAM 2 | |
|
|
| --- |
|
|
| ## ⚡ Quickstart: Python & ONNX Runtime |
|
|
| ```python |
| import onnxruntime as ort |
| import numpy as np |
| from PIL import Image |
| |
| # 1. Initialize U-Net Inference Session |
| session = ort.InferenceSession("unet_model.onnx", providers=["CPUExecutionProvider"]) |
| |
| # 2. Preprocess Input Image |
| image = Image.open("sample_carpet.jpg").convert("RGB").resize((512, 512)) |
| img_data = np.array(image, dtype=np.float32) / 255.0 |
| img_data = np.transpose(img_data, (2, 0, 1)) |
| img_data = np.expand_dims(img_data, axis=0) |
| |
| # 3. Execute Model Prediction |
| input_name = session.get_inputs()[0].name |
| output_name = session.get_outputs()[0].name |
| prediction = session.run([output_name], {input_name: img_data})[0] |
| |
| # 4. Generate Binary Mask |
| segmentation_mask = (prediction[0, 0] > 0.5).astype(np.uint8) * 255 |
| print("Inference successful. Mask dimension:", segmentation_mask.shape) |
| ``` |
|
|
| --- |
|
|
| ## 👤 Author |
| * **Developer:** [yosedie](https://huggingface.co/yosedie) |
| * **GitHub:** [https://github.com/yosedie](https://github.com/yosedie) |
|
|
