How asasori/pbl5 works
Main Components
A. Data Processing Layer
• Input Processing (Camera/Video Stream)
• Frame Processing (30-frame and realtime processing paths)
• Data Collection and Preprocessing (indicated by Ipynb/collect_data_optional.ipynb)
B. Model Architecture Layer
Multiple model implementations should be shown:
• 1D CNN + Transformer Models (various versions)
• LSTM Models
• Uniformer Models
C. Streaming System
• Real-time streaming components
• 30-frame streaming components
• Client-Server Architecture
[Input Layer] → [Processing Layer] → [Inference Layer] → [Output Layer]
• ┌─ 1D CNN + Transformer Branch
• LSTM Branch
• Uniformer Branch
Key Elements to Highlight
• Blue: Input/Output components
• Green: Processing components
• Yellow: Model components
• Gray: Support services
• Rectangles: Processing units
• Diamonds: Decision points
Important Architectural Details
• Client-server architecture for both streaming implementations:
• [Client] ←→ [Server Handler] ←→ [Server]
• Model versioning system:
• [Model Repository]
• Train 8
• Train 9
• Train 10...13
Implementation Notes
• Version numbers for different model implementations
• Frame processing paths (30-frame vs. realtime)
Technical Specifications
• Model architectures (1DCNN_Transformer, LSTM, Uniformer)
• Frame processing specifications (15F, 30F)
• Dimension information (dim192, dim256, etc.)
• Input/Output formats