Table 8.

Parameters for training the CascadeTabNet model

Description	Value
Input image size (height x width)	1024x1024
Backbone model used for feature extraction	ResNet-50
Number of output channels in the last layer of the backbone	256
Number of input channels	256
Number of fully connected (FC) layers	2
Number of output channels for each FC layer	1024
Number of stages in the cascade	3
Region Proposal Network (RPN) output threshold	2000
RPN minimum positive IoU threshold	0.3
Number of object classes (table or background)	2
Learning rate of the optimizer	0.005
Loss function for classification	Cross Entropy
Loss function for bounding box regression	Smooth L1