--- Log opened Sat Feb 20 00:00:34 2021
01:20 -!- darsie [~kvirc@84-113-55-200.cable.dynamic.surfer.at] has joined ##hplusroadmap
01:23 < TMA> on the other hand, there is usually reduced ambiguity between 0oO, Il1|, rn/m and such when you take the pdfs in a non-image way
01:34 < fenn> you could feed the OCR a dictionary from the PDF text
04:02 -!- sanehatter [~sanehatte@141.98.255.147] has quit [Quit: -]
04:04 -!- biglot00 [~biglot00@212.102.45.109] has joined ##hplusroadmap
04:06 -!- biglot00 [~biglot00@212.102.45.109] has left ##hplusroadmap []
04:17 -!- sanehatter [sanehatter@gateway/vpn/mullvad/sanehatter] has joined ##hplusroadmap
04:19 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0:5cda:ee35:7600:7b2a] has joined ##hplusroadmap
05:47 -!- docl_ is now known as docl
08:24 -!- CryptoDavid [uid14990@gateway/web/irccloud.com/x-xsfyarruxqtezbep] has joined ##hplusroadmap
09:46 -!- Malvolio [~Malvolio@unaffiliated/malvolio] has quit [Quit: brb]
09:51 -!- Malvolio [~Malvolio@unaffiliated/malvolio] has joined ##hplusroadmap
10:11 -!- geekitgood [~Adium@2600-6c67-6e7f-18d6-d417-4003-7e43-76ae.res6.spectrum.com] has joined ##hplusroadmap
10:12 -!- geekitgood1 [~Kaleb@2600-6c67-6e7f-18d6-9c50-97b9-4c6c-71ae.res6.spectrum.com] has joined ##hplusroadmap
10:17 -!- geekitgood1 [~Kaleb@2600-6c67-6e7f-18d6-9c50-97b9-4c6c-71ae.res6.spectrum.com] has left ##hplusroadmap []
10:20 -!- geekitgood1 [~Kaleb@2600-6c67-6e7f-18d6-9c50-97b9-4c6c-71ae.res6.spectrum.com] has joined ##hplusroadmap
10:30 -!- geekitgood [~Adium@2600-6c67-6e7f-18d6-d417-4003-7e43-76ae.res6.spectrum.com] has quit [Quit: Leaving.]
10:35 -!- geekitgood [~Adium@2600-6c67-6e7f-18d6-d417-4003-7e43-76ae.res6.spectrum.com] has joined ##hplusroadmap
10:37 -!- geekitgood [~Adium@2600-6c67-6e7f-18d6-d417-4003-7e43-76ae.res6.spectrum.com] has left ##hplusroadmap []
10:57 -!- geekitgood1 [~Kaleb@2600-6c67-6e7f-18d6-9c50-97b9-4c6c-71ae.res6.spectrum.com] has quit [Quit: Leaving.]
11:27 -!- spaceangel [~spaceange@ip-94-112-205-34.net.upcbroadband.cz] has joined ##hplusroadmap
12:16 < lsneff> I'd bet a combined approach would be best. Use ocr to determine what elements correspond to what, and then chain together the text elements actually present at those locations in the pdf.
14:24 -!- geekitgood [~Kaleb@2600-6c67-6e7f-18d6-0cde-7a8f-7c5c-3b95.res6.spectrum.com] has joined ##hplusroadmap
15:13 -!- preview [~quassel@2407:7000:8423:b00::3] has quit [Ping timeout: 240 seconds]
15:19 -!- spaceangel [~spaceange@ip-94-112-205-34.net.upcbroadband.cz] has quit [Remote host closed the connection]
15:23 -!- SDr [~SDr@unaffiliated/sdr] has quit [Ping timeout: 256 seconds]
15:24 -!- SDr [~SDr@unaffiliated/sdr] has joined ##hplusroadmap
15:35 < pasky> fenn: since justanotheruser is talking about CNNs he likely wants to work on rendered PDF (which is absolutely the right call)
15:36 < pasky> unfortunately i don't have up-to-date idea about opensource tools in this domain; we have moved pretty far from low-level developer tool to an end-to-end business solution meanwhile
16:20 -!- CryptoDavid [uid14990@gateway/web/irccloud.com/x-xsfyarruxqtezbep] has quit [Quit: Connection closed for inactivity]
16:58 -!- filipepe_ [uid362247@gateway/web/irccloud.com/x-otsyqjgmofpkydkz] has joined ##hplusroadmap
17:01 -!- geekitgood [~Kaleb@2600-6c67-6e7f-18d6-0cde-7a8f-7c5c-3b95.res6.spectrum.com] has left ##hplusroadmap []
17:04 -!- join_cordblood [~join_cord@135-23-248-163.cpe.pppoe.ca] has quit [Ping timeout: 256 seconds]
17:36 -!- join_cordblood [~join_cord@135-23-248-163.cpe.pppoe.ca] has joined ##hplusroadmap
17:58 -!- darsie [~kvirc@84-113-55-200.cable.dynamic.surfer.at] has quit [Ping timeout: 240 seconds]
18:56 -!- yashgaroth [~ffffffff@2601:5c4:c780:6aa0:5cda:ee35:7600:7b2a] has quit [Quit: Leaving]
19:04 -!- SDr [~SDr@unaffiliated/sdr] has quit [Ping timeout: 265 seconds]
19:15 -!- SDr [~SDr@unaffiliated/sdr] has joined ##hplusroadmap
20:17 -!- preview [~quassel@2407:7000:8423:b03:8f:7e45:612d:862b] has joined ##hplusroadmap
21:18 -!- filipepe_ [uid362247@gateway/web/irccloud.com/x-otsyqjgmofpkydkz] has quit [Quit: Connection closed for inactivity]
21:29 -!- juri_ [~juri@178.63.35.222] has quit [Ping timeout: 272 seconds]
22:09 -!- SDr [~SDr@unaffiliated/sdr] has quit [Ping timeout: 272 seconds]
22:16 -!- SDr [~SDr@unaffiliated/sdr] has joined ##hplusroadmap
22:28 < justanotheruser> In theory a CNN could probably work with raw pdf, but that seams suboptimal :p
22:31 < justanotheruser> also wouldn't have the flexibility of being applicable to other documents
22:32 < justanotheruser> lsneff: I've tried that and it results in a massive load of edge case handling in the dataset I'm working with.
22:32 -!- thahxa [~thahxa@vlnsm7-toronto63-142-116-127-110.internet.virginmobile.ca] has joined ##hplusroadmap
22:33 < lsneff> Honestly, the easiest way that might work well enough would probably be to do shitty ocr on the pdf, and then dump that through gpt3 and ask it to sort everything into a nice, labeled list.
--- Log closed Sun Feb 21 00:00:35 2021