Spaces:
Runtime error
Runtime error
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,171 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
|
2 |
+
|
3 |
+
|
4 |
+
import tensorflow as tf
|
5 |
+
from matplotlib import pyplot as plt
|
6 |
+
from skimage.transform import rescale, resize
|
7 |
+
import pickle as pkl
|
8 |
+
import numpy as np
|
9 |
+
import os
|
10 |
+
import cv2
|
11 |
+
from PIL import Image,ImageFont, ImageDraw
|
12 |
+
import CALTextModel
|
13 |
+
|
14 |
+
|
15 |
+
|
16 |
+
#### training setup parameters ####
|
17 |
+
|
18 |
+
|
19 |
+
lambda_val=1e-4
|
20 |
+
gamma_val=1
|
21 |
+
|
22 |
+
|
23 |
+
|
24 |
+
|
25 |
+
|
26 |
+
################################### Utility functions###################################
|
27 |
+
|
28 |
+
|
29 |
+
|
30 |
+
def load_dict_picklefile(dictFile):
|
31 |
+
fp=open(dictFile,'rb')
|
32 |
+
lexicon=pkl.load(fp)
|
33 |
+
fp.close()
|
34 |
+
return lexicon,lexicon[' ']
|
35 |
+
|
36 |
+
def preprocess_img(img):
|
37 |
+
if len(img.shape)>2:
|
38 |
+
img= cv2.cvtColor(img.astype('float32'), cv2.COLOR_BGR2GRAY)
|
39 |
+
height=img.shape[0]
|
40 |
+
width=img.shape[1]
|
41 |
+
|
42 |
+
if(width<300):
|
43 |
+
result = np.ones([img.shape[0], img.shape[1]*2])*255
|
44 |
+
result[0:img.shape[0],img.shape[1]:img.shape[1]*2]=img
|
45 |
+
img=result
|
46 |
+
|
47 |
+
img=cv2.resize(img, dsize=(800,100), interpolation = cv2.INTER_AREA)
|
48 |
+
img=(img-img.min())/(img.max()-img.min())
|
49 |
+
xx_pad = np.zeros((100, 800), dtype='float32')
|
50 |
+
xx_pad[:,:] =1
|
51 |
+
|
52 |
+
xx_pad = xx_pad[None, :, :]
|
53 |
+
img=img[None, :, :]
|
54 |
+
return img, xx_pad
|
55 |
+
|
56 |
+
|
57 |
+
worddicts,_ = load_dict_picklefile('vocabulary.pkl')
|
58 |
+
worddicts_r = [None] * len(worddicts)
|
59 |
+
i=1
|
60 |
+
for kk, vv in worddicts.items():
|
61 |
+
if(i<len(worddicts)):
|
62 |
+
worddicts_r[vv] = kk
|
63 |
+
else:
|
64 |
+
break
|
65 |
+
i=i+1
|
66 |
+
|
67 |
+
# Create an instance of the model
|
68 |
+
CALTEXT = CALTextModel.CALTEXT_Model(training=False)
|
69 |
+
CALTEXT.load_weights('final_caltextModel/cp-0037.ckpt')
|
70 |
+
test_loss = tf.keras.metrics.Mean(name='test_loss')
|
71 |
+
|
72 |
+
|
73 |
+
@tf.function(experimental_relax_shapes=True)
|
74 |
+
def execute_model(xx,xx_mask,CALTEXT):
|
75 |
+
|
76 |
+
anno = CALTEXT(xx,xx_mask, training=False)
|
77 |
+
hidden_state_0 = CALTEXT.get_hidden_state_0(anno)
|
78 |
+
return anno,hidden_state_0
|
79 |
+
|
80 |
+
|
81 |
+
def test_error( images, x_mask):
|
82 |
+
# training=False is only needed if there are layers with different
|
83 |
+
# behavior during training versus inference (e.g. Dropout).
|
84 |
+
batch_loss=0
|
85 |
+
img_ind=1
|
86 |
+
for img_ind in range(len(images)):
|
87 |
+
xx = images[img_ind][tf.newaxis, ... ]
|
88 |
+
xx_mask = x_mask[img_ind][tf.newaxis, ... ]
|
89 |
+
anno,hidden_state_0=execute_model(xx,xx_mask,CALTEXT)
|
90 |
+
|
91 |
+
sample, score,hypalpha=CALTextModel.get_sample(anno, hidden_state_0,10, 130, False, False, CALTEXT)
|
92 |
+
|
93 |
+
|
94 |
+
score = score / np.array([len(s) for s in sample])
|
95 |
+
ss = sample[score.argmin()]
|
96 |
+
img_ind=img_ind+1
|
97 |
+
|
98 |
+
ind=0
|
99 |
+
num=int(len(ss)/2)
|
100 |
+
|
101 |
+
|
102 |
+
|
103 |
+
|
104 |
+
#### output string
|
105 |
+
ind=0
|
106 |
+
outstr=u''
|
107 |
+
frames = []
|
108 |
+
#font = ImageFont.truetype("Jameel Noori Nastaleeq.ttf",60)
|
109 |
+
while (ind<len(ss)-1):
|
110 |
+
k=(len(ss)-2)-ind
|
111 |
+
outstr=outstr+worddicts_r[int(ss[k])]
|
112 |
+
'''textimg = Image.new('RGB', (1400,100),(255,255,255))
|
113 |
+
drawtext = ImageDraw.Draw(textimg)
|
114 |
+
drawtext.text((20, 20), outstr ,(0,0,0),font=font)
|
115 |
+
fig,axes=plt.subplots(2,1)
|
116 |
+
axes[0].imshow(textimg)
|
117 |
+
axes[0].axis('off')
|
118 |
+
axes[1].axis('off')
|
119 |
+
axes[1].imshow(xx[0,:,:],cmap='gray')
|
120 |
+
visualization=resize(hypalpha[k], (100,800),anti_aliasing=True)
|
121 |
+
axes[1].imshow(255-(255 * visualization), alpha=0.2)
|
122 |
+
plt.axis('off')
|
123 |
+
|
124 |
+
plt.savefig('/content/gdrive/My Drive/CALText_Demo/res.png')
|
125 |
+
frames.append(Image.fromarray(cv2.imread('/content/gdrive/My Drive/CALText_Demo/res.png'), 'RGB'))'''
|
126 |
+
ind=ind+1
|
127 |
+
'''frame_one = frames[0]
|
128 |
+
frame_one.save("/content/gdrive/My Drive/CALText_Demo/'vis.gif", format="GIF", append_images=frames,save_all=True, duration=300, loop=0)
|
129 |
+
gif_image="/content/gdrive/My Drive/CALText_Demo/'vis.gif"'''
|
130 |
+
return outstr,gif_image
|
131 |
+
|
132 |
+
|
133 |
+
|
134 |
+
''''''examples = [
|
135 |
+
['/content/gdrive/My Drive/CALText_Demo/sample_test_images/59-11.png'],
|
136 |
+
['/content/gdrive/My Drive/CALText_Demo/sample_test_images/59-21.png'],
|
137 |
+
['/content/gdrive/My Drive/CALText_Demo/sample_test_images/59-32.png'],
|
138 |
+
['/content/gdrive/My Drive/CALText_Demo/sample_test_images/59-37.png'],
|
139 |
+
['/content/gdrive/My Drive/CALText_Demo/sample_test_images/91-47.png'],
|
140 |
+
['/content/gdrive/My Drive/CALText_Demo/sample_test_images/91-49.png'],
|
141 |
+
]''''''
|
142 |
+
|
143 |
+
|
144 |
+
import gradio as gr
|
145 |
+
|
146 |
+
def recognize_text(input_image):
|
147 |
+
x, x_mask=preprocess_img(input_image)
|
148 |
+
output_str,gif_image=test_error(x, x_mask)
|
149 |
+
return output_str,gif_image
|
150 |
+
|
151 |
+
|
152 |
+
title = "CALText Demo"
|
153 |
+
description = "<p style='text-align: center'>Gradio demo for an CALText model architecture <a href='https://github.com/nazar-khan/CALText'>[GitHub Code]</a> trained on the <a href='http://faculty.pucit.edu.pk/nazarkhan/work/urdu_ohtr/pucit_ohul_dataset.html'>PUCIT-OHUL</a> dataset. To use it, simply add your image, or click one of the examples to load them. </p>"
|
154 |
+
article = "<p style='text-align: center'></p>"
|
155 |
+
css = "#0 {object-fit: contain;} #1 {object-fit: contain;}"
|
156 |
+
inputs = gr.inputs.Image(label="Input Image")
|
157 |
+
|
158 |
+
demo = gr.Interface(fn=recognize_text,inputs=inputs,outputs=[gr.Textbox(label="Output"),gr.Image(label="Demonstration of attention")],title=title,
|
159 |
+
description=description,
|
160 |
+
article=article,allow_flagging='never')
|
161 |
+
|
162 |
+
demo.launch()
|
163 |
+
|
164 |
+
|
165 |
+
|
166 |
+
|
167 |
+
|
168 |
+
|
169 |
+
|
170 |
+
|
171 |
+
|