PaddleOCR - probably the best OCR tool available

[tuzi]

2021.10.03

• Update PaddleOCR.dll to 2.3.

• Update detection model to ch_PP-OCRv2_det.

• Update ImagePut.ahk to 1.2 beta.

• Load configs no longer requires temporary file.

• Rename dict file and model directories

[leosouza85]

I'm using ocr clipboard
printscreen and the script, but its does not ocr everything on the screen, why?

[leosouza85]

It is possible to use this script to find a word and then click it?

[tuzi]

Thanks for sharing
Why can't I copy content to the clipboard?

Code: Select all

o := PaddleOCR(ClipboardAll)
MsgBox, % o
Clipboard := o

Even when I press CTRL +C on the msgbox, it causes it to disappear

Copying such a msgbox will get its contents and will not cause it to disappear

Code: Select all

o := "my text"
MsgBox, % o

i don't have this problem, maybe you can try version 2021.10.03 ?

[tuzi]

I'm using ocr clipboard
printscreen and the script, but its does not ocr everything on the screen, why?

look 0.note.ahk

then set "visualize":1 in your code.

[tuzi]

It is possible to use this script to find a word and then click it?

look 6.Get all information.ahk

use the pos of str. then click it.

[leosouza85]

It is possible to use this script to find a word and then click it?

look 6.Get all information.ahk

use the pos of str. then click it.

Thank you so much!

[tuzi]

It is possible to use this script to find a word and then click it?

look 6.Get all information.ahk

use the pos of str. then click it.

Thank you so much!

this tool FindText maybe better for you.

[leosouza85]

It is possible to use this script to find a word and then click it?

look 6.Get all information.ahk

use the pos of str. then click it.

Thank you so much!

this tool FindText maybe better for you.

Yes, I know this tool, but I'm looking for new and flexible approaches, any tips to increase the accuracy on PaddleOCR?

[tuzi]

Try to adjust the configs, for example

det_db_thresh
det_db_box_thresh
det_db_unclip_ratio

@leosouza85

[arcticir]

Sorry. Edited.

[malcev]

arcticir, thanks for info, but why in chinese?

[sanmaodo]

@tuzi
Thank you from the bottom of my heart！

[hughman]

only support x64? is it possible for x86?

[leosouza85]

I'm using ocr clipboard
printscreen and the script, but its does not ocr everything on the screen, why?

Answering myself but in hope that will help others, I've changed the PaddleOCR.ahk parameter max_side_len, from 960 to 1920 that is my horizontal resolution, and now it OCR the whole screen.

[tuzi]

2021.11.24

• When set "get_all_info" to 1, the return value is an object.

• Fix a bug when set "get_all_info" to 1.

• Update example 6 and 7.

• Update ImagePut.ahk to 1.3.

[iseahound]

One thing that I noticed is that if the file name and the window name are the same, "text.png" is open in a window with the same title - it will use the window over the file.

I don't know if this bothers anyone.

[tuzi]

2022.01.10

• Support OCR PDF file directly.

• Fix a bug when set "get_all_info" to 1.

• Add 4 examples.

• Update all examples.

• Update ImagePut.ahk to 1.5.1.

[newcod3r]

Thank you for this.

1. May I know how to train the model to get better? it is still quite inaccurate based on this screenshot:

image.png (137.56 KiB) Viewed 3190 times

2. How do I get ahk to select the region first before it inputs the coordinates into the function? A sample will be appreciated. Thanks!

[tuzi]

Thank you for this.

1. May I know how to train the model to get better? it is still quite inaccurate based on this screenshot:

image.png

2. How do I get ahk to select the region first before it inputs the coordinates into the function? A sample will be appreciated. Thanks!

1. https://github.com/PaddlePaddle/PaddleOCR

2. You're lucky I ever wrote that.(you need btt lib or replace all btt() to tooltip)

Code: Select all

arr := select_range()

; MsgBox, % PaddleOCR(arr)

select_range()
{
  KeyWait, LButton
  For k,v in StrSplit("LButton,Left,Right,Up,Down", ",")
    Hotkey, % v, GetRange_KeyOff, On
  if (0)
  {
    GetRange_KeyOff:
    return
  }
  Gui, GetRange: Destroy
  Gui, % "GetRange: +AlwaysOnTop -Caption +ToolWindow"
    . " +Hwndid -DPIScale +E0x08000000"
  ok:=0
  CoordMode, Mouse, Screen
  While !GetKeyState("Ctrl","P")
  {
    Sleep, 50
    if (ok!=2)
      btt("drag to select a range",,, 2, {FontSize:14})
    MouseGetPos, x2, y2, win
    if (ok=0)
    {
      if GetKeyState("LButton","P")
        ok:=1, x1:=x2, y1:=y2, outWin:=win
    }
    else if (ok=1)
    {
      GetKeyState("Left","P")  && x1--
      GetKeyState("Right","P") && x1++
      GetKeyState("Up","P")    && y1--
      GetKeyState("Down","P")  && y1++
      w:=Floor(Abs(x1-x2)), h:=Floor(Abs(y1-y2))
      x:=Floor(Min(x1,x2)), y:=Floor(Min(y1,y2))
      d:=2, x-=d, y-=d, w+=2*d, h+=2*d, i:=w-d, j:=h-d
      s:="0-0 " (w "-0 ") (w "-" h) (" 0-" h) " 0-0  "
      . (d "-" d) " " (i "-" d) " " (i "-" j) " " (d "-" j) " " (d "-" d)
      WinSet, Region, %s%, ahk_id %id%
      Gui, GetRange: Color, % (A_MSec<500 ? "Red":"Blue")
      Gui, GetRange: Show, NA x%x% y%y% w%w% h%h%
      if !GetKeyState("LButton","P")
        ok:=2
    }
    else if (ok=2)
    {
      btt("arrow keys to fine-tune the position`nclick the left button to confirm",,, 2, {FontSize:14})
      GetKeyState("Left","P")  && x--
      GetKeyState("Right","P") && x++
      GetKeyState("Up","P")    && y--
      GetKeyState("Down","P")  && y++
      Gui, GetRange: Color, % (A_MSec<500 ? "Red":"Blue")
      Gui, GetRange: Show, NA x%x% y%y%
      if GetKeyState("LButton","P")
        Break
    }
  }
  x1:=x+d, y1:=y+d, x2:=x+w-1-d, y2:=y+h-1-d
  Gui, GetRange: Destroy
  btt(,,, 2)
  KeyWait, LButton
  For k,v in StrSplit("LButton,Left,Right,Up,Down", ",")
    Hotkey, % v, Off
  return [x1, y1, Abs(x2-x1), Abs(y2-y1), outWin]
}