Spaces:
Running
Running
add online demo
Browse files- .gitignore +1 -0
- app.py +96 -4
- label.names +84 -0
- model.tflite +3 -0
- samples/00.jpg +0 -0
- samples/01.jpg +0 -0
- samples/02.jpg +0 -0
- samples/03.jpg +0 -0
- samples/04.jpg +0 -0
- samples/06.jpg +0 -0
.gitignore
ADDED
@@ -0,0 +1 @@
|
|
|
|
|
1 |
+
.DS_Store
|
app.py
CHANGED
@@ -1,7 +1,99 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2 |
|
3 |
-
def greet(name):
|
4 |
-
return "Hello " + name + "!!"
|
5 |
|
6 |
-
|
7 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import glob
|
3 |
+
from itertools import groupby
|
4 |
+
|
5 |
+
import cv2
|
6 |
+
import numpy as np
|
7 |
import gradio as gr
|
8 |
+
import tensorflow as tf
|
9 |
+
|
10 |
+
|
11 |
+
def get_sample_images():
|
12 |
+
list_ = glob.glob(os.path.join(os.path.dirname(__file__), 'samples/*.jpg'))
|
13 |
+
return [[i] for i in list_]
|
14 |
+
|
15 |
+
|
16 |
+
def inference(image):
|
17 |
+
# load model
|
18 |
+
demo = TFliteDemo(os.path.join(os.path.dirname(__file__), 'model.tflite'))
|
19 |
+
# load image
|
20 |
+
image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
|
21 |
+
image = center_fit(image, 128, 64, top_left=True)
|
22 |
+
image = np.reshape(image, (1, *image.shape, 1)).astype(np.uint8)
|
23 |
+
# inference
|
24 |
+
pred = demo.inference(image)
|
25 |
+
# decode
|
26 |
+
dict = load_dict(os.path.join(os.path.dirname(__file__), 'label.names'))
|
27 |
+
res = decode_label(pred, dict)
|
28 |
+
return res
|
29 |
+
|
30 |
+
|
31 |
+
class TFliteDemo:
|
32 |
+
def __init__(self, model_path):
|
33 |
+
self.interpreter = tf.lite.Interpreter(model_path=model_path)
|
34 |
+
self.interpreter.allocate_tensors()
|
35 |
+
self.input_details = self.interpreter.get_input_details()
|
36 |
+
self.output_details = self.interpreter.get_output_details()
|
37 |
+
|
38 |
+
def inference(self, x):
|
39 |
+
self.interpreter.set_tensor(self.input_details[0]['index'], x)
|
40 |
+
self.interpreter.invoke()
|
41 |
+
return self.interpreter.get_tensor(self.output_details[0]['index'])
|
42 |
+
|
43 |
+
|
44 |
+
def center_fit(img, w, h, inter=cv2.INTER_NEAREST, top_left=True):
|
45 |
+
# get img shape
|
46 |
+
img_h, img_w = img.shape[:2]
|
47 |
+
# get ratio
|
48 |
+
ratio = min(w / img_w, h / img_h)
|
49 |
+
|
50 |
+
if len(img.shape) == 3:
|
51 |
+
inter = cv2.INTER_AREA
|
52 |
+
# resize img
|
53 |
+
img = cv2.resize(img, (int(img_w * ratio), int(img_h * ratio)), interpolation=inter)
|
54 |
+
# get new img shape
|
55 |
+
img_h, img_w = img.shape[:2]
|
56 |
+
# get start point
|
57 |
+
start_w = (w - img_w) // 2
|
58 |
+
start_h = (h - img_h) // 2
|
59 |
+
|
60 |
+
if top_left:
|
61 |
+
start_w = 0
|
62 |
+
start_h = 0
|
63 |
+
|
64 |
+
if len(img.shape) == 2:
|
65 |
+
# create new img
|
66 |
+
new_img = np.zeros((h, w), dtype=np.uint8)
|
67 |
+
new_img[start_h:start_h+img_h, start_w:start_w+img_w] = img
|
68 |
+
else:
|
69 |
+
new_img = np.zeros((h, w, 3), dtype=np.uint8)
|
70 |
+
new_img[start_h:start_h+img_h, start_w:start_w+img_w, :] = img
|
71 |
+
|
72 |
+
return new_img
|
73 |
+
|
74 |
+
|
75 |
+
def load_dict(dict_path='label.names'):
|
76 |
+
with open(dict_path, 'r', encoding='utf-8') as f:
|
77 |
+
dict = f.read().splitlines()
|
78 |
+
dict = {i: dict[i] for i in range(len(dict))}
|
79 |
+
return dict
|
80 |
+
|
81 |
+
|
82 |
+
def decode_label(mat, chars) -> str:
|
83 |
+
# mat is the output of model
|
84 |
+
# get char indices along best path
|
85 |
+
best_path_indices = np.argmax(mat[0], axis=-1)
|
86 |
+
# collapse best path (using itertools.groupby), map to chars, join char list to string
|
87 |
+
best_chars_collapsed = [chars[k] for k, _ in groupby(best_path_indices) if k != len(chars)]
|
88 |
+
res = ''.join(best_chars_collapsed)
|
89 |
+
return res
|
90 |
|
|
|
|
|
91 |
|
92 |
+
interface = gr.Interface(
|
93 |
+
fn=inference,
|
94 |
+
inputs="image",
|
95 |
+
outputs="text",
|
96 |
+
title="South Korean License Plate Recognition",
|
97 |
+
examples=get_sample_images(),
|
98 |
+
)
|
99 |
+
interface.launch()
|
label.names
ADDED
@@ -0,0 +1,84 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
0
|
2 |
+
1
|
3 |
+
2
|
4 |
+
3
|
5 |
+
4
|
6 |
+
5
|
7 |
+
6
|
8 |
+
7
|
9 |
+
8
|
10 |
+
9
|
11 |
+
가
|
12 |
+
나
|
13 |
+
다
|
14 |
+
라
|
15 |
+
마
|
16 |
+
거
|
17 |
+
너
|
18 |
+
더
|
19 |
+
러
|
20 |
+
머
|
21 |
+
버
|
22 |
+
서
|
23 |
+
어
|
24 |
+
저
|
25 |
+
고
|
26 |
+
노
|
27 |
+
도
|
28 |
+
로
|
29 |
+
모
|
30 |
+
보
|
31 |
+
소
|
32 |
+
오
|
33 |
+
조
|
34 |
+
구
|
35 |
+
누
|
36 |
+
두
|
37 |
+
루
|
38 |
+
무
|
39 |
+
부
|
40 |
+
수
|
41 |
+
우
|
42 |
+
주
|
43 |
+
하
|
44 |
+
허
|
45 |
+
호
|
46 |
+
바
|
47 |
+
사
|
48 |
+
아
|
49 |
+
자
|
50 |
+
배
|
51 |
+
서울
|
52 |
+
부산
|
53 |
+
대구
|
54 |
+
인천
|
55 |
+
광주
|
56 |
+
대전
|
57 |
+
울산
|
58 |
+
세종
|
59 |
+
경기
|
60 |
+
강원
|
61 |
+
충북
|
62 |
+
충남
|
63 |
+
전북
|
64 |
+
전남
|
65 |
+
경북
|
66 |
+
경남
|
67 |
+
제주
|
68 |
+
북
|
69 |
+
제
|
70 |
+
종
|
71 |
+
전
|
72 |
+
충
|
73 |
+
천
|
74 |
+
울
|
75 |
+
원
|
76 |
+
인
|
77 |
+
대
|
78 |
+
세
|
79 |
+
산
|
80 |
+
남
|
81 |
+
강
|
82 |
+
기
|
83 |
+
경
|
84 |
+
광
|
model.tflite
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:a16ab2c9446d3edfa5e620ae418ed68952dd0047d8656eba2001a269341c0f81
|
3 |
+
size 307488
|
samples/00.jpg
ADDED
samples/01.jpg
ADDED
samples/02.jpg
ADDED
samples/03.jpg
ADDED
samples/04.jpg
ADDED
samples/06.jpg
ADDED