Form detection and recognition

CascadeTabNet：An Approach for end to end table detection and structure recognition from image-based documents(2020)

Paper
DevashishPrasad/CascadeTabNet

• Network structure
  
• Iterative transfer learning
  uses multi-task learning methods and multiple iterations of transfer learning to achieve goals. First train the model on a general data set, and then fine-tune multiple times for a specific data set.
  
• Image Transformation and data augmentation (image transformation and data augmentation)
  • Dilation transform
    In the dilation transform, the original image is first converted into a binary image, and then the original image is transformed to thicken the black pixel area.
  • Smudge transform
    
• Pipeline
  
• result