3 ms·

This task is on my TODO list. Why does http://calibre-ebook.com/ http://calibre-ebook.com/ fail?

by keefe 17y ago

roel_v 17y ago
Because it is basically a glorified frontend (as far as pdf to other formats conversion is concerned) for pdftohtml. It treats every line as a paragraph, can't intelligently decide on what is a chapter heading etc. The problem I'm describing is AI hard, it can't be done be some rule based matching. I know that Calibre has some regex based rules for finding chapters etc. but that is way too hard and only solves part of the problem, besides many book don't use the 'Chapter 1: xxx' format anyway.
- keefe 17y ago
  We may have different goals in mind... I have a bunch of PDFs I need to get to my sony reader, for example academic papers. I want them to look basically OK, which I don't "think" is an AI hard problem. Detecting any text that represents a chapter certainly is... but I'm OK with writing regex to get my data out. I'll be posting in the mobileread forums about this at some point in the coming weeks if you're interested in whatever solution I hack out.
  - roel_v 17y ago
    Academic papers especially are a PITA, at least those formatted in columns, so it depends on the journal I guess. Plus you need detection of paragraphs to get line breaks at the proper locations. If it's not in columns and if you are willing to tweak the algorithm parameters to each conversion (paper), then it won't be that hard (just detect increased line spacing, or indented first line), but a commercial application would have to work almost automatically. or otherwise have a very quick feedback cycle (preview/edit settings/redo conversion). That's another one of my gripes with Calibre, the epub/fb2 preview application is so so slow that it's almost unusable. Yeah let me know if you find a working solution, although I do have dozens of hours worth of scripts laying about to do custom conversions, so real one-off hackjobs I have so many of that I don't need any more :)

This task is on my TODO list. Why does http://calibre-ebook.com/ http://calibre-ebook.com/ fail?

by keefe 17y ago

This task is on my TODO list. Why does http://calibre-ebook.com/ http://calibre-ebook.com/ fail?

Because it is basically a glorified frontend (as far as pdf to other formats conversion is concerned) for pdftohtml. It treats every line as a paragraph, can't intelligently decide on what is a chapter heading etc. The problem I'm describing is AI hard, it can't be done be some rule based matching. I know that Calibre has some regex based rules for finding chapters etc. but that is way too hard and only solves part of the problem, besides many book don't use the 'Chapter 1: xxx' format anyway.

keefe 17y ago
We may have different goals in mind... I have a bunch of PDFs I need to get to my sony reader, for example academic papers. I want them to look basically OK, which I don't "think" is an AI hard problem. Detecting any text that represents a chapter certainly is... but I'm OK with writing regex to get my data out. I'll be posting in the mobileread forums about this at some point in the coming weeks if you're interested in whatever solution I hack out.
- roel_v 17y ago
  Academic papers especially are a PITA, at least those formatted in columns, so it depends on the journal I guess. Plus you need detection of paragraphs to get line breaks at the proper locations. If it's not in columns and if you are willing to tweak the algorithm parameters to each conversion (paper), then it won't be that hard (just detect increased line spacing, or indented first line), but a commercial application would have to work almost automatically. or otherwise have a very quick feedback cycle (preview/edit settings/redo conversion). That's another one of my gripes with Calibre, the epub/fb2 preview application is so so slow that it's almost unusable. Yeah let me know if you find a working solution, although I do have dozens of hours worth of scripts laying about to do custom conversions, so real one-off hackjobs I have so many of that I don't need any more :)