How burakulas/detocs works
Main Components
a) Web Interface Layer:
• User input handling for single/multiple target processing
b) Processing Layer:
• Utils module as a supporting component
c) Model Layer:
• SSD (Single Shot Detector)
• Faster R-CNN
• EfficientDet D1
• To a model selection switch
d) Data Layer:
• Components for:
• Asset storage (labelmap files)
• Training data collection
• TESS EBS external data source
Component Relationships
a) Data Flow:
• User input → Web Interface
• Web Interface → Processing Layer
• Processing Layer ↔ Model Layer
• Processing Layer → Data Layer (for training data collection)
• External TESS EBS → Processing Layer
b) Model Selection Flow:
• How the confidence threshold affects model selection
• Model switching capability
Important Details to Highlight
• A note about GPU requirement for Faster R-CNN and EfficientDet
• Confidence threshold input affecting model performance
• Training data collection feedback loop
• Period factor calculation process
Technical Specifications
• Supported input formats
• Output format [class, x-center, y-center, width, height]
• Supported file types for batch processing
Training data collection
Remember to indicate the asynchronous nature of the web interface and the processing pipeline, and show how the system handles multiple concurrent requests.