PP Table Classification and Detection Pipeline¶

This pipeline uses deep learning models and lightweight heuristics for Table Structure Recognition. It combines table classification with specialized cell detection:

1. Classify each table as wired or wireless
2. Detect cells in the table using appropriate set of weights for the class
3. Reconstruct row and column structure with positional heuristics

graph LR
  A[Table image] --> B{PP-LCNet_x1_0_table_cls};
  B -->|wired| C[RT-DETR-L_wired_table_cell_det];
  B -->|wireless| D[RT-DETR-L_wireless_table_cell_det];
  C --> E[Cell detections];
  D --> E[Cell detections];
  E -->|Heuristics| F[Table structure]

This pipeline is provided as a Docling plugin for being part of a complete document parsing pipeline. Check our documentation on usage with Docling.