some questions about char2wordBB and the generated text

1. I don't understand the function char2wordBB, what's the meaning of "cc = np.squeeze(np.concatenate(np.dsplit(cc,cc.shape[-1]),axis=1)).T.astype('float32')" and " cc_tblr"? Someone can help me?

2. I met some cases with the generated text.

1. The text is underlined. "know" in the first pic.
2. "." and text overlap. "rec.sport.hockey" in the second pic.
3. the bounding box may appear to be not tightly enclosed. "27" and “Apr”in the third pic.
4. word is not rendered successfully. "files" and "that" in the fourth pic.