Latexocr generate completely off result

[shrninepoints]

The pix2tex command line tool works really well for me, generating high quality results most of the time. But the GUI by calling latexocr gives completely random results.

E.g. when pix2tex generate $E=m c^{2}$ the GUI generate $\scriptstyle{\hat{s e}}_{k\in G a l}^{a=1,10}$

Environment: macOS 12.6.3, Python 3.7.7, PyQt5

PS, can you add option in pix2tex command line to automatically bracket with $$? Thanks!

[lukas-blecher]

Can you tell me more about your setup. Do you use a second monitor? Do you have a retina display?

I'm still wondering how exactly the screen capture works on macOS. The way you explained the problem, it sounds like the region you select is not actually the region the model gets to see.
I can't really do anything about that at the moment, because I don't have a mac book.

If you want you can find the gui.py file and add img.show() in line 300 to see what the model will see.

https://github.com/lukas-blecher/LaTeX-OCR/blob/main/pix2tex/gui.py#L300

like this:

                img = ImageGrab.grab(bbox=(x1//factor, y1//factor, x2//factor, y2//factor), all_screens=True)
            else:
                raise e
+      img.show()
        QApplication.processEvents()

Possibly related #122 #118 #145

[shrninepoints]

Thank you for the quick response. It's macbook pro 2018 with retina display. I didn't use a second monitor in the reported case. I tried adding img.show() as suggested, it turns out the captured image is a random part of my desktop wallpaper, even if the whole desktop is covered by some other window. The position of the captured image in the wallpaper is not related to where I actually capture it. When I actually use a second monitor, latexocr gives the following error message """"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" gui.py 297 mouseReleaseEvent img = ImageGrab.grab(bbox=(x1//factor, y1//factor, x2//factor, y2//factor), all_screens=True) ImageGrab.py 38 grab im = Image.open(filepath) Image.py 3124 open "cannot identify image file %r" % (filename if filename else fp) PIL.UnidentifiedImageError: cannot identify image file '/var/folders/q7/xqj1xs_13ms9y9327t0t28sc0000gn/T/tmpqnne9nm7.png' screencapture: -R requires a valid rect (x,y,w,h) screencapture: -R requires a valid rect (x,y,w,h) """"""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""""" The file tmpqnne9nm7.png is an empty file. In some cases the file path is '/tmp/tmpqnne9nm7.png'. Anyways thank you for the wonderful tool and good luck with your development! Best, H Lukas Blecher ***@***.***> 于2023年2月17日周五 23:34写道：

…

Can you tell me more about your setup. Do you use a second monitor? Do you have a retina display? I'm still wondering how exactly the screen capture works on macOS. The way you explained the problem, it sounds like the region you select is not actually the region the model gets to see. I can't really do anything about that at the moment, because I don't have a mac book. If you want you can find the gui.py file and add img.show() in line 300 to see what the model will see. https://github.com/lukas-blecher/LaTeX-OCR/blob/main/pix2tex/gui.py#L300 like this: img = ImageGrab.grab(bbox=(x1//factor, y1//factor, x2//factor, y2//factor), all_screens=True) else: raise e+ img.show() QApplication.processEvents() Possibly related #122 <#122> #118 <#118> #145 <#145> — Reply to this email directly, view it on GitHub <#241 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHWNVSADOG344T3WPEUIKCTWX6K6XANCNFSM6AAAAAAU7RS6MU> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[kaaass]

According to BoboTiG/python-mss#134, Grant screen record permission to the Terminal app (I use the Terminal app to run latexocr) solved the problem for me.

[skymagician]

According to BoboTiG/python-mss#134, Grant screen record permission to the Terminal app (I use the Terminal app to run latexocr) solved the problem for me.

thank you! I met the same problem, and solved it according to your method.