"tav output modes": typo! I meant to say "TSV output mode". Sorry.
Met vriendelijke groeten / Best regards, Ger Hobbelt -------------------------------------------------- web: http://www.hobbelt.com/ http://www.hebbut.net/ mail: [email protected] mobile: +31-6-11 120 978 -------------------------------------------------- On Wed, 5 Nov 2025, 09:01 Ger Hobbelt, <[email protected]> wrote: > Sorry, can't help further. Like I said before: this reads as having to run > this in a debugger and see what happens. > > What DOES jump into the eye are those very odd (HUGE) b-box coordinate > numbers: what you would expect to be X/y pixel coordinates of the original > image and /nobody/ has images with over a billion pixels in the horizontal > axis! All those 4 numbers are suspect, which leads me to suspect the binary > API interface between go and c++ is possibly broken. No certainty but this > smells pretty bad. > > For reference and to aid your debugging efforts, go and see what tesseract > cli outputs re X/y coordinates in hocr of tav output modes. The bbox > numbers should fall in the same price range, so to speak. ;-) > > > Met vriendelijke groeten / Best regards, > > Ger Hobbelt > > -------------------------------------------------- > web: http://www.hobbelt.com/ > http://www.hebbut.net/ > mail: [email protected] > mobile: +31-6-11 120 978 > -------------------------------------------------- > > On Mon, 3 Nov 2025, 14:47 Harshit Goel, <[email protected]> wrote: > >> Hi Ger, >> >> Thanks a lot for the detailed guidance — it was really helpful. >> >> I ran deeper diagnostics and confirmed a few things: >> >> - >> >> Running *Tesseract CLI* directly works perfectly and extracts: *NO >> SMOKING* >> - >> >> However, when using *gosseract* from Go, I still get *empty text >> output* and a single empty bounding box like: >> Text: [ ], Box: (1476397136,32579)-(1476956064,32579) >> >> The image being processed is a valid 8-bit/16-bit PNG (confirmed via file >> command). >> >> - >> >> Setting *TESSDATA_PREFIX *or >> *SetTessdataPrefix("/usr/share/tessdata")* works correctly — no >> language load errors. >> - >> >> Even after forcing engine mode with *tessedit_ocr_engine_mode = 1 *(LSTM >> only) and using *PSM_SPARSE_TEXT*, gosseract still returns empty text. >> - >> >> This makes me think gosseract is initializing Tesseract differently >> (maybe not loading the same configs or missing something in the setup >> phase), because the CLI and Go layer are using the same image and >> tessdata. >> >> Do you have any suggestions for checking whether gosseract is properly >> initializing TessBaseAPI with the same defaults as CLI? >> >> Thanks again for your help — your earlier hint about checking bounding >> boxes and configuration alignment was spot on. >> >> Best regards, >> Harshit >> >> On Sun, Nov 2, 2025 at 4:09 AM Ger Hobbelt <[email protected]> wrote: >> >>> I expect you're in for a debug session. >>> >>> I do not use Go, so here's just a few general tidbits: >>> >>> - you tested with the tesseract CLI. Excellent! So that proves things >>> can go well at the core; one major problem area less to worry about. >>> - next is the gosseract library/layer itself: how does it talk to >>> tesseract, what does it pass (and what doesn't it), etc.: from a very swift >>> glance at the code, there's nothing blatantly obviously wrong in their >>> bindings.cpp, AFAICT. Haven;t looked any further than that. >>> - my own usage of tesseract as a library has shown me that getting the >>> parameters right can be a bit of a hassle sometimes; one of the potential >>> failure modes is not noting that tesseract does not receive the same config >>> baseline setup as when it ran via CLI: this is where debugging is mandatory. >>> >>> My first guess would be to make very sure your tesseract config files >>> are loaded the same way. While that can be a bit harsh to do when you're >>> not comfortable with running this stuff in a debugger, here's a preparation >>> step I would definitely look at if I were you: >>> 1. tesseract via your Go code doesn't produce *anything*, while >>> 2. tesseract CLI does deliver text ("No smoking") >>> which MAY be due to tesseract not finding any text word bounding-boxes >>> when run via the Go-code route. >>> >>> I see they (gosseract) present a GetBoundingBoxes API, so I would first >>> try to run that one to see if I get any boxes at all, and if any, where >>> they are in the image (i.e.: do I get: (a) no boxes, (b) only get gibberish >>> boxes only or (c) at least the ones covering "NO" and "SMOKING", or what? >>> Then try the same for the CLI (IIRC vanilla tesseract has an option to >>> cough up bboxes only; haven't used that in a while and I'm running a >>> customized tesseract here, so check code and documentation, don't take me >>> at my word!) >>> >>> To see what I was looking at: >>> https://github.com/otiai10/gosseract/blob/main/tessbridge.cpp#L108 >>> >>> If the bounding boxes don't show up in your Go run, then it smells like >>> a config/setup bit not making it into the tesseract engine, so it's >>> debugging the gosseract bindings.cpp interlayer to see what happens, >>> really. Are CLI and Go code really, really pointing at the same config >>> search paths, for example? >>> If the bounding boxes show up and match the set in the CLI, we have a >>> serious conundrum. >>> >>> Either way, that's the road I'd travel if walking in your shoes. >>> (If you can debug-step the tesseract CLI the same way, you can more >>> easily compare both, perhaps, as the CLI is using the same APIs gosseract >>> is using (with some differences, but my current bet is those are not >>> relevant). >>> >>> Also monitor the gosseract/tesseract run for error and warning messages >>> from tesseract, as well. If it is silent, maybe force it once to barf a >>> hairball, just so you know the error/warning/info outputs are working. >>> Whatever you do, my bet is you have some debugging on the road ahead. >>> >>> Note: I don't do Go, so haven't used gosseract. This would be my general >>> tactic though, anyway. >>> >>> >>> Met vriendelijke groeten / Best regards, >>> >>> Ger Hobbelt >>> >>> -------------------------------------------------- >>> web: http://www.hobbelt.com/ >>> http://www.hebbut.net/ >>> mail: [email protected] >>> mobile: +31-6-11 120 978 >>> -------------------------------------------------- >>> >>> >>> On Fri, Oct 31, 2025 at 6:45 PM Harshit Goel <[email protected]> >>> wrote: >>> >>>> Hi team >>>> >>>> I’m facing an issue where Tesseract OCR works correctly from the CLI, >>>> but returns an empty string when called programmatically using Go (via >>>> gosseract). >>>> >>>> For this particular image: >>>> https://pmi-api.ubconnex.ca/files/icons/2025-03/11c6051eec503f52c43f0de382980d31.png, >>>> the OCR always returns an empty string when running programmatically. Yet >>>> when I run the exact same image manually using Tesseract from terminal by >>>> command: *tesseract /tmp/ocr-3678469497.png stdout* >>>> >>>> It correctly detects and returns *NO SMOKING* >>>> >>>> *Environment* >>>> >>>> - OS: Linux (Server) >>>> - >>>> >>>> Tesseract version: tesseract 5.x (CLI works fine) >>>> - >>>> >>>> Go binding: github.com/otiai10/gosseract/v2 >>>> - >>>> >>>> Go version: go1.23.x >>>> >>>> I've tried with the following approaches but still no effect: >>>> >>>> - >>>> >>>> Different PSM modes (SPARSE_TEXT, SINGLE_BLOCK, etc.) >>>> - >>>> >>>> Preprocessing (grayscale, contrast enhancement, flattening >>>> transparency). >>>> - >>>> >>>> Verified that the image file is saved correctly and readable by >>>> Tesseract. >>>> - >>>> >>>> Tried increasing image size and contrast. >>>> >>>> Is there any known discrepancy between the CLI binary and the gosseract >>>> API in how page segmentation modes or image preprocessing are handled >>>> internally? >>>> >>>> Any insight on why Tesseract detects text in CLI but gosseract binding >>>> returns empty output would be very helpful. >>>> >>>> Best Regards, >>>> >>>> Harshit Goel >>>> >>>> -- >>>> You received this message because you are subscribed to the Google >>>> Groups "tesseract-ocr" group. >>>> To unsubscribe from this group and stop receiving emails from it, send >>>> an email to [email protected]. >>>> To view this discussion visit >>>> https://groups.google.com/d/msgid/tesseract-ocr/54875e13-9f91-4f45-9eb8-ee8eec4e5846n%40googlegroups.com >>>> <https://groups.google.com/d/msgid/tesseract-ocr/54875e13-9f91-4f45-9eb8-ee8eec4e5846n%40googlegroups.com?utm_medium=email&utm_source=footer> >>>> . >>>> >>> -- >>> You received this message because you are subscribed to the Google >>> Groups "tesseract-ocr" group. >>> To unsubscribe from this group and stop receiving emails from it, send >>> an email to [email protected]. >>> To view this discussion visit >>> https://groups.google.com/d/msgid/tesseract-ocr/CAFP60foBhh_8kWyiP9-zVyfO8JrxwgDmvm%3DZH5pnE3sHYiu_1g%40mail.gmail.com >>> <https://groups.google.com/d/msgid/tesseract-ocr/CAFP60foBhh_8kWyiP9-zVyfO8JrxwgDmvm%3DZH5pnE3sHYiu_1g%40mail.gmail.com?utm_medium=email&utm_source=footer> >>> . >>> >> -- >> You received this message because you are subscribed to the Google Groups >> "tesseract-ocr" group. >> To unsubscribe from this group and stop receiving emails from it, send an >> email to [email protected]. >> To view this discussion visit >> https://groups.google.com/d/msgid/tesseract-ocr/CADRW4UeJiWeZa6aO%2BS2pZoqG1zkMX0q18Rg0efCk7irb5u6Zsw%40mail.gmail.com >> <https://groups.google.com/d/msgid/tesseract-ocr/CADRW4UeJiWeZa6aO%2BS2pZoqG1zkMX0q18Rg0efCk7irb5u6Zsw%40mail.gmail.com?utm_medium=email&utm_source=footer> >> . >> > -- You received this message because you are subscribed to the Google Groups "tesseract-ocr" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion visit https://groups.google.com/d/msgid/tesseract-ocr/CAFP60fp-DTer%2BMRBssiWeeOxHbTe8fJxYpi14QcwuXWvaH-3WQ%40mail.gmail.com.
