The following major code changes have been made,

• Order of function arguments changed as per relevance (tensors become the first argument and later other hyperparameters).
• conv_normal_initializer() removed from function arguments as it is directly accessible inside the calling function.
• Function names have been renamed/shortened.
• Multiline input arguments to functions compressed inside args/kwargs.
• compute_MBconv_block_parameters() moved outside of process_feature_maps()
• Most of the arguments to ClassNet() and BoxNet() are common and hence put under args.
• Layer names completely removed.
• get_drop_connect() optimized.
• Unit test for get_drop_connect() added.
• All function docstrings compressed.
• All the codes converted to pep8 style.

The model trains well and produces outputs as expected.
All test cases pass.