Skip to content
Projects
Groups
Snippets
Help
This project
Loading...
Sign in / Register
Toggle navigation
tutorials
Overview
Overview
Details
Activity
Cycle Analytics
Repository
Repository
Files
Commits
Branches
Tags
Contributors
Graph
Compare
Charts
Issues
0
Issues
0
List
Board
Labels
Milestones
Merge Requests
0
Merge Requests
0
CI / CD
CI / CD
Pipelines
Jobs
Schedules
Charts
Wiki
Wiki
Snippets
Snippets
Members
Members
Collapse sidebar
Close sidebar
Activity
Graph
Charts
Create a new issue
Jobs
Commits
Issue Boards
Open sidebar
Data Science Dojo
tutorials
Commits
4e1d64d7
Commit
4e1d64d7
authored
Apr 22, 2021
by
Usman Shahid
Browse files
Options
Browse Files
Download
Email Patches
Plain Diff
Upload New File
parent
0f1bd8f2
Hide whitespace changes
Inline
Side-by-side
Showing
1 changed file
with
323 additions
and
0 deletions
+323
-0
functions.py
Querying Azure Custom Vision in Python/functions.py
+323
-0
No files found.
Querying Azure Custom Vision in Python/functions.py
0 → 100644
View file @
4e1d64d7
import
os
# for converting the frames into bytes
import
cv2
# and for processing arrays
import
numpy
as
np
# for encoding and decoding Custom Vision predictions
import
json
# for converting the Custom Vision predictions to dataframe
import
pandas
as
pd
# import async packages
import
asyncio
import
aiohttp
# for file name pattern matching
import
fnmatch
# for displaying images from the processes output video
import
matplotlib.pyplot
as
plt
# importing other required libraries
import
random
import
textwrap
import
datetime
from
PIL
import
Image
import
time
import
GlobalVariables
# web service end-point for the Custom Vision model
# we will process video frames (which are images)
POST_URL
=
"Your custom vision model endpoint"
# providing prediction key
HEADERS
=
{
'Prediction-Key'
:
"Your custom vision model prediction key"
,
"Content-Type"
:
"application/json"
}
# number of API calls per pool of request
MAX_CONNECTIONS
=
100
# initializing the height and width for frames in the video
WIDTH
=
0
HEIGHT
=
0
# returns frame array of video
def
getFrames
(
path
):
images
=
[]
byteImages
=
[]
vidObj
=
cv2
.
VideoCapture
(
path
)
count
=
0
success
=
1
while
success
:
success
,
image
=
vidObj
.
read
()
if
success
:
images
.
append
(
image
)
count
+=
1
global
HEIGHT
,
WIDTH
HEIGHT
,
WIDTH
,
lay
=
images
[
0
]
.
shape
return
images
# Returns ByteImages for POST request params
def
convertCVImagesToByteImages
(
frames
):
toReturnByteImages
=
[]
for
frame
in
frames
:
image
=
cv2
.
cvtColor
(
frame
,
cv2
.
COLOR_BGR2RGB
)
is_success
,
im_buf_arr
=
cv2
.
imencode
(
".jpg"
,
image
)
if
is_success
:
byte_im
=
im_buf_arr
.
tobytes
()
toReturnByteImages
.
append
(
byte_im
)
return
toReturnByteImages
# making async requests call
async
def
fetch
(
session
,
url
,
data
,
i
):
print
(
"Request #"
+
str
(
i
)
+
" Sent"
)
async
with
session
.
post
(
url
,
headers
=
HEADERS
,
data
=
data
)
as
response
:
resp
=
await
response
.
text
()
print
(
"Request #"
+
str
(
i
)
+
" Complete"
)
return
resp
# Setting up for batch requests
async
def
fetch_all
(
byteImages
):
conn
=
aiohttp
.
TCPConnector
(
limit
=
len
(
byteImages
)
+
1
)
async
with
aiohttp
.
ClientSession
(
connector
=
conn
)
as
session
:
tasks
=
[]
i
=
0
for
image
in
byteImages
:
i
=
i
+
1
# if i%20 == 0:
# time.sleep(3)
tasks
.
append
(
fetch
(
session
,
POST_URL
,
image
,
i
)
)
responses
=
await
asyncio
.
gather
(
*
tasks
,
return_exceptions
=
True
)
return
responses
# defining a function to convert raw JSON to a dataframe
def
getDFfromResponse
(
response
,
threshold
):
resp
=
json
.
loads
(
response
)
pred
=
resp
[
'predictions'
]
pred_df
=
pd
.
DataFrame
(
pred
)
resp_df
=
pred_df
[
pred_df
[
'probability'
]
>
float
(
threshold
)]
return
resp_df
# defining a function to converting the bounding box dataframe to a list
def
boundingBoxListFromDF
(
df
):
cood_list
=
df
[
'boundingBox'
]
.
tolist
()
return
cood_list
# defining a function for adjusting the size of bounding box
def
fitAccordingToSize
(
coods
):
print
(
"Multiplying by "
+
str
(
WIDTH
)
+
" , "
+
str
(
HEIGHT
))
for
i
in
range
(
len
(
coods
)):
coods
[
i
][
'left'
]
=
coods
[
i
][
'left'
]
*
WIDTH
coods
[
i
][
'top'
]
=
coods
[
i
][
'top'
]
*
HEIGHT
coods
[
i
][
'width'
]
=
coods
[
i
][
'width'
]
*
WIDTH
coods
[
i
][
'height'
]
=
coods
[
i
][
'height'
]
*
WIDTH
return
coods
#==================== function: makeRequests ====================#
# defining a function to get bounding box coordinates and tags
# makes calls to other functions defined above
def
makeRequests
(
byteImages
,
threshold
):
tagsData
=
[]
bound_box_cood
=
[]
responseJSON
=
[]
# futures = [...]
loop
=
asyncio
.
get_event_loop
()
check
=
loop
.
run_until_complete
(
fetch_all
(
byteImages
))
# check = asyncio.run()
# check = await fetch_all(byteImages)
for
i
,
response
in
enumerate
(
check
):
df
=
getDFfromResponse
(
response
,
threshold
)
listBx
=
boundingBoxListFromDF
(
df
)
listBx
=
fitAccordingToSize
(
listBx
)
taglist
=
df
[
'tagName'
]
.
tolist
()
tagsData
.
append
(
taglist
)
bound_box_cood
.
append
(
listBx
)
responseJSON
.
append
(
df
)
return
bound_box_cood
,
tagsData
,
responseJSON
# defining a function to check compliance or non-compliance for a frame
def
isCompliant
(
tags
):
return
(
len
(
list
(
filter
(
lambda
x
:
x
==
'mask'
,
tags
)))
>=
len
(
list
(
filter
(
lambda
x
:
x
==
'person'
,
tags
))))
# Print for Compliance Label
def
PrintCompliance
(
frames
,
i
,
font
,
lineType
):
frames
[
i
]
=
cv2
.
rectangle
(
frames
[
i
],
(
WIDTH
-
300
,
0
),
(
WIDTH
,
50
),
(
0
,
128
,
0
),
-
1
)
frames
[
i
]
=
cv2
.
putText
(
frames
[
i
],
"COMPLIANT"
,
(
WIDTH
-
220
,
33
),
font
,
1
,
(
255
,
255
,
255
),
2
,
lineType
)
# Print for Non-Compliance Label
def
PrintNonCompliance
(
frames
,
i
,
font
,
lineType
):
frames
[
i
]
=
cv2
.
rectangle
(
frames
[
i
],
(
WIDTH
-
300
,
0
),
(
WIDTH
,
50
),
(
0
,
0
,
128
),
-
1
)
frames
[
i
]
=
cv2
.
putText
(
frames
[
i
],
"NON-COMPLIANT"
,
(
WIDTH
-
275
,
34
),
font
,
1
,
(
255
,
255
,
255
),
2
,
lineType
)
#==================== function: printCNC ====================#
#Check for compliance or non-compliance label
def
printCNC
(
which
,
frames
,
i
,
font
,
lineType
):
if
which
:
PrintCompliance
(
frames
,
i
,
font
,
lineType
)
else
:
PrintNonCompliance
(
frames
,
i
,
font
,
lineType
)
# defining a function to convert a list of frames into a video
def
SaveVideo
(
frames
,
output_file_name
):
img_array
=
[]
for
img
in
frames
:
height
,
width
,
layers
=
img
.
shape
size
=
(
width
,
height
)
img_array
.
append
(
img
)
out
=
cv2
.
VideoWriter
(
output_file_name
,
cv2
.
VideoWriter_fourcc
(
*
'MP4V'
),
15
,
size
)
for
i
in
range
(
len
(
img_array
)):
out
.
write
(
img_array
[
i
])
out
.
release
()
# defining a function to convert raw input video to processed video with tags and stats
def
ConvertVideo
(
video_path
,
output_file_name
,
threshold
,
nframes
):
frames
=
getFrames
(
video_path
)
# this converts the input video into frames
if
nframes
!=
'all'
:
frames
=
frames
[:
int
(
nframes
)]
byteImages
=
convertCVImagesToByteImages
(
frames
)
# this converts frames into list of byte images
coodsData
,
tagsData
,
responses
=
makeRequests
(
byteImages
,
threshold
)
# this returns lists of coordinate data, tag data, responses in dataframe
font
=
cv2
.
FONT_HERSHEY_SIMPLEX
bottomLeftCornerOfText
=
(
50
,
50
)
fontScale
=
0.5
fontColor
=
(
255
,
255
,
255
)
lineType
=
1
frames_stats
=
[]
for
i
,
coods_single
in
enumerate
(
coodsData
):
dflist
=
responses
[
i
]
.
drop
(
'boundingBox'
,
axis
=
1
)
print
(
dflist
[
'tagId'
])
dflist
[
'tagId'
]
=
(
dflist
[
'tagId'
]
.
to_string
())[
-
10
:]
dflist
.
reset_index
(
drop
=
True
,
inplace
=
True
)
d
=
dflist
.
to_dict
(
orient
=
'records'
)
jo
=
json
.
dumps
(
d
)
width
=
600
height
=
1080
blank_image
=
np
.
zeros
((
height
,
width
,
3
),
np
.
uint8
)
n_masks
=
len
([
x
for
x
in
tagsData
[
i
]
if
x
==
'mask'
])
n_people
=
len
([
x
for
x
in
tagsData
[
i
]
if
x
==
'person'
])
color
=
(
0
,
0
,
0
)
blank_image
[:
height
//
2
]
=
color
blank_image
[
height
//
2
:]
=
(
0
,
0
,
0
)
blank_image
=
cv2
.
putText
(
blank_image
,
"STATISTICS"
,(
230
,
70
),
font
,
1
,(
66
,
245
,
173
),
2
,
lineType
)
blank_image
=
cv2
.
rectangle
(
blank_image
,
(
80
,
150
),
(
width
-
60
,
220
),
(
66
,
245
,
173
),
2
)
blank_image
=
cv2
.
rectangle
(
blank_image
,
(
80
,
260
),
(
width
-
60
,
330
),
(
66
,
245
,
173
),
2
)
blank_image
=
cv2
.
putText
(
blank_image
,
"NO OF PEOPLE: "
+
str
(
n_people
),(
90
+
20
,
195
),
font
,
0.9
,(
66
,
245
,
173
),
2
,
lineType
)
blank_image
=
cv2
.
putText
(
blank_image
,
"NO OF MASKS: "
+
str
(
n_masks
),(
90
+
20
,
305
),
font
,
0.9
,(
66
,
245
,
173
),
2
,
lineType
)
relative_start
=
height
//
2
+
-
150
x
,
y0
=
(
50
,
relative_start
)
text
=
dflist
.
to_string
()
thickness
=
10
text_size
,
_
=
cv2
.
getTextSize
(
text
,
font
,
fontScale
,
thickness
)
line_height
=
text_size
[
1
]
+
20
if
len
(
text
)
!=
0
:
for
k
,
line
in
enumerate
(
text
.
split
(
"
\n
"
)):
y
=
y0
+
k
*
line_height
relative_start
=
y
if
k
==
0
:
ctext
=
' Probability Id Tag'
blank_image
=
cv2
.
putText
(
blank_image
,
ctext
,
(
x
,
y
),
font
,
0.8
,
(
66
,
245
,
173
),
2
,
lineType
)
else
:
blank_image
=
cv2
.
putText
(
blank_image
,
line
,
(
x
,
y
),
font
,
0.8
,
(
66
,
245
,
173
),
2
,
lineType
)
wrapped_text
=
textwrap
.
wrap
(
jo
,
width
=
50
)
x
,
y0
=
(
10
,
relative_start
)
if
relative_start
>
700
:
y
=
relative_start
else
:
y
=
700
for
p
,
line
in
enumerate
(
wrapped_text
):
print
(
line
)
y
=
y
+
30
blank_image
=
cv2
.
putText
(
blank_image
,
line
,
(
10
,
y
),
font
,
0.7
,
(
66
,
245
,
173
),
2
,
lineType
)
frames_stats
.
append
(
blank_image
)
last
=
None
if
len
(
tagsData
[
i
])
==
0
:
pass
elif
not
isCompliant
(
tagsData
[
i
]):
# this checks for compliance for the frame
if
i
%
5
==
0
:
PrintNonCompliance
(
frames
,
i
,
font
,
lineType
)
# this prints non-compliant labels
last
=
0
else
:
if
last
==
None
:
PrintNonCompliance
(
frames
,
i
,
font
,
lineType
)
last
=
0
else
:
printCNC
(
last
,
frames
,
font
,
lineType
)
# this checks for compliance before applying compliance labels
else
:
if
i
%
5
==
0
:
PrintCompliance
(
frames
,
i
,
font
,
lineType
)
# this prints compliant labels
last
=
1
else
:
if
last
==
None
:
PrintCompliance
(
frames
,
i
,
font
,
lineType
)
last
=
1
else
:
printCNC
(
last
,
frames
,
font
,
lineType
)
for
j
,
single
in
enumerate
(
coods_single
):
# this puts the tab on top of the detected item
frames
[
i
]
=
cv2
.
rectangle
(
frames
[
i
],
(
int
(
single
[
'left'
])
-
3
,
int
(
single
[
'top'
])
-
30
),
((
int
(
single
[
'left'
])
+
100
,
int
(
single
[
'top'
])
-
3
)),
(
66
,
245
,
173
),
-
1
)
frames
[
i
]
=
cv2
.
putText
(
frames
[
i
],
tagsData
[
i
][
j
],
(
int
(
single
[
'left'
]
+
2
),
int
(
single
[
'top'
])
+
-
8
),
font
,
0.9
,
(
0
,
0
,
0
),
2
,
lineType
)
# this creates the bounding box
frames
[
i
]
=
cv2
.
rectangle
(
frames
[
i
],
(
int
(
single
[
'left'
]),
int
(
single
[
'top'
])),
(
int
(
single
[
'left'
])
+
int
(
single
[
'width'
]),
int
(
single
[
'top'
])
+
int
(
single
[
'height'
])),
(
66
,
245
,
173
)
,
3
)
frames
[
i
]
=
np
.
concatenate
((
frames
[
i
],
frames_stats
[
i
]),
axis
=
1
)
SaveVideo
(
frames
,
output_file_name
)
# this converts frames into a video
stats_path
=
output_file_name
.
replace
(
'Output'
,
'Stats'
)
SaveVideo
(
frames_stats
,
stats_path
)
\ No newline at end of file
Write
Preview
Markdown
is supported
0%
Try again
or
attach a new file
Attach a file
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Cancel
Please
register
or
sign in
to comment