How mwasifanwar/neurofabric-ai works
Project Type and Purpose
• neurofabric-ai is a backend ML service for real-time computer vision on textile production lines. It has no frontend GUI; instead it exposes an HTTP API for:
• Defect detection in live fabric streams
• Fabric analysis metrics
• Generative design of new fabric patterns via GANs
High-Level Architecture
• External Clients (Manufacturing Cameras, Monitoring Dashboards)
• API Layer (FastAPI or Flask in run_api.py/main.py)
• Core Processing Pipeline (fabric_analyzer → defect_detector → quality_controller → design_generator)
• Model Store (CNN and GAN model definitions & weights)
• Utility Services (image preprocessing, data loading, visualization)
• Configuration (model parameters, settings)
• Tests (unit & integration)
Main Components & Their Relationships
a. API Layer
– Entry point: run_api.py / main.py
– Routes: api/routes.py
– Middleware (authentication, logging): api/middleware.py
– Delegates requests to Core Processing Pipeline
b. Core Processing Pipeline
– fabric_analyzer: extracts metrics from raw images
– defect_detector: applies CNN models to find defects
– quality_controller: aggregates results, enforces thresholds
– design_generator: uses GAN models to produce new designs
c. Model Store
– cnn_models.py, gan_models.py define architecture and load weights
– config/model_config.py specifies paths, hyperparameters
d. Utilities
– data_loader.py: ingest streaming or batch image data
– image_processor.py: resize, normalize, augment images
– visualization.py: render detection overlays or design previews
e. Configuration
– config/settings.py: global settings (e.g., API ports, log levels)
– config/model_config.py: model-specific settings
f. Tests
– tests/test_defect_detector.py, test_fabric_analyzer.py, test_integration.py validate functionality
Key Architectural Patterns
• Layered Architecture: clear separation between API, core business logic, model layer, and utils
• Pipeline Pattern: sequential processing stages for image data
• Config-Driven: decouple hard-coded parameters into config modules
• Modular ML: separate CNN vs. GAN submodules for reuse and testing
Final Checklist
• Annotations for frameworks & libraries