The key difference is the document model. PDF describes where every glyph sits on a sheet; FB2 describes structure: chapters, paragraphs, headings, images. During conversion we extract the PDF's text layer, re-join words split by line-wrap hyphens, group lines back into paragraphs, and rebuild the chapter split. Chapter boundaries come from the PDF's table of contents (bookmarks) when present; otherwise chapters are detected from large headings in the text. The result reads in any e-reader just like a book bought from a store.
The tool has a few conveniences. The title and author are filled in automatically from the PDF properties – you just check them instead of typing (empty fields are fine too). You can pick a genre for correct cataloguing in your reader's library. Every image from the PDF is embedded straight into the FB2 in reading order, so the book opens offline. By default the first page becomes the cover – handy when the PDF starts with a title page; the cover can be turned off.
One limitation: FB2 is built only from a text-based PDF. If your file is a scan of a paper book (image-only pages with no text layer), the tool will say so plainly. In that case run the OCR tool first to get a PDF with a text layer, then convert it to FB2. The resulting .fb2 opens in all popular readers: FBReader, PocketBook, CoolReader, AlReader, and the desktop Calibre.