ksasso1028/easyocr-cpp

Custom C++ implementation of deep learning based OCR

Welcome to my experiments 👋

Independent ML consultant working across computer vision, audio ML DSP, and biosignals

Recently, I designed and shipped the AI declipper which just released in FL Studio 21.1 - > credit

Also a active musician,and continuing ML research into audio restoration/generation, video generation, and classification of biosignals.

contact -> [email protected]

easyocr-cpp's Issues

How do you export the EasyOCR models to TorchScript format?

I would like to deploy EasyOCR models using libtorch, and saw your repo does this, which is excellent.

But the official EasyOCR models are not in TorchScript format needed by this software.

Do you have a conversion script or similar?

Specifically, I am having a hard time exporting the recognizer model - I don't know how to create the input tensor.

Best regards,
Mikael

linux needs change in CmakeList.txt

Hi,
in order to compile in linux you need to add
set(CMAKE_CXX_STANDARD 20)

dockerization

Hello,
Thanks for your repo! You really did what you said there.
Some suggestions that I have is that your repo is too much dependent on your local computer configuration, I thought a Dockerfile to play with your repo would be nice to have.
If you think it is valuable to add it I can add my Dockerfile, it automatically gets all dependencies: get libtorch and build latest OpenCV and build the application.

Also I've tried your code with get but got an error, seems that some tensor are in CPU while other are in GPU. Let's dig this point later
I would have some C++ specific comments on the code but let it for later, a PR was requested to solve a linux build.

Is it possible to use only the designated model?

std::string record = This part is not available even if you include language models from other countries.
And do not use the English model (.pth) provided by easyocr.

output do not make sense in Linux

Hi,
could you please help me to understand what is happening? This is the output I obtain when I run it in linux. The text detected completely messed up.

0 [blank]
95 z
94 y
93 x
92 w
91 v
90 u
89 t
88 s
87 r
86 q
85 p
84 o
83 n
82 m
81 l
80 k
79 j
78 i
77 h
76 g
75 f
74 e
73 d
72 c
71 b
70 a
69 Z
68 Y
67 X
66 W
65 V
64 U
63 T
62 S
61 R
60 Q
29 =
28 <
27 ;
26 :
25 /
24 .
23 -
22 ,
21 +
20 *
19 )
18 (
17 '
16 &
15 %
14 $
1 0
2 1
3 2
4 3
5 4
6 5
7 6
8 7
9 8
10 9
11 !
12 "
13 #
30 >
31 ?
32 @
33 [
34
35 ]
36 ^
37 _
38 `
39 {
40 |
41 }
42 ~
43
44 A
45 B
46 C
47 D
48 E
49 F
50 G
51 H
52 I
53 J
54 K
55 L
56 M
57 N
58 O
59 P
MODEL TIME 45.7036 ms
LOCATION: [0, 1] [20, 20]
TEXT: Q
CONFIDENCE 0.999953
################################################
LOCATION: [21, 5] [68, 18]
TEXT: IPMPOJ
CONFIDENCE 0.999209
################################################
LOCATION: [71, 5] [108, 18]
TEXT: UISFF
CONFIDENCE 0.999888
################################################
LOCATION: [115, 5] [349, 18]
TEXT: 490054792 Pqfofe cw QXIpmpo NEDbsejp
CONFIDENCE 0.995813
################################################
LOCATION: [13, 29] [38, 40]
TEXT: Ubdl
CONFIDENCE 0.999953
################################################
LOCATION: [53, 29] [78, 42]
TEXT: Feju
CONFIDENCE 0.999371
################################################
LOCATION: [91, 29] [118, 40]
TEXT: Kjfm
CONFIDENCE 0.0257791
################################################
LOCATION: [135, 29] [176, 42]
TEXT: Qbujfou
CONFIDENCE 0.999988
################################################
LOCATION: [191, 29] [222, 42]
TEXT: Dibsu
CONFIDENCE 0.999963
################################################
LOCATION: [237, 29] [266, 40]
TEXT: Mjolf
CONFIDENCE 0.999889
################################################
LOCATION: [281, 29] [353, 40]
TEXT: Opujgjdbujpot
CONFIDENCE 0.982506
################################################
LOCATION: [367, 29] [427, 42]
TEXT: Obwjrbujpo
CONFIDENCE 0.996567
################################################
LOCATION: [443, 29] [473, 44]
TEXT: Ifmq
CONFIDENCE 0.998373
################################################
LOCATION: [15, 55] [28, 68]
TEXT: J
CONFIDENCE 0.010791
################################################
LOCATION: [29, 55] [64, 66]
TEXT: Ipnf
CONFIDENCE 0.987937
################################################
LOCATION: [89, 55] [168, 66]
TEXT: Dbsejpwbtdvmbs
CONFIDENCE 0.999805
################################################
LOCATION: [177, 55] [186, 68]
TEXT: Jb
CONFIDENCE 0.00381894
################################################
LOCATION: [191, 55] [377, 70]
TEXT: Qbujfou Mjtu Qfsjpqfsbujwf Usbdljoh
CONFIDENCE 0.999996
################################################
LOCATION: [401, 55] [603, 66]
TEXT: Qspwjefs Iboepgg F Nfttbef Dfoufs
CONFIDENCE 0.999982
################################################
LOCATION: [624, 55] [722, 66]
TEXT: Ewobnjd Xpslmjtu
CONFIDENCE 0.999048
################################################
LOCATION: [742, 55] [858, 68]
TEXT: Sfgfssbm Nbobhfnfou
CONFIDENCE 0.997839
################################################
LOCATION: [882, 55] [958, 68]
TEXT: NzFyqfsjfodf
CONFIDENCE 0.999368
################################################
LOCATION: [15, 81] [28, 94]
TEXT: M
CONFIDENCE 0.999945
################################################
LOCATION: [29, 79] [86, 94]
TEXT: VqUpEbuf
CONFIDENCE 0.999511
################################################
LOCATION: [29, 107] [56, 120]
TEXT: EBa
CONFIDENCE 0.999826
################################################
LOCATION: [13, 133] [28, 146]
TEXT: XpMb
CONFIDENCE 0.0275535
################################################
LOCATION: [35, 133] [74, 144]
TEXT: Qspop
CONFIDENCE 0.999905
################################################
LOCATION: [89, 133] [130, 144]
TEXT: Lcopo
CONFIDENCE 0.990057

How to ise GPU?

Hello, nice project. How can I run it with GPU support (have already installed lobtorch with cuda special version that I have on pc)?

output text doesn't match with actual image

I run the torchTest program with input is test.jpg in the repo but the output doesn't match with actual image. Each character is shifted to right one unit.
Ex: text in image is THREE, output is UISFF
T -> U
H -> I
R -> S
E -> F
E -> F

How to OCR Chinese Characters

Hi,
How to OCR Chinese Characters,I do not know how to do,Can you give me a suggestion?
Thanks!