Spaces:
Sleeping
Sleeping
Commit
•
8dafa03
1
Parent(s):
d020d74
Upload 3 files
Browse files- app.py +116 -0
- demo.pkl +3 -0
- requirements.txt +10 -0
app.py
ADDED
@@ -0,0 +1,116 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import streamlit as st
|
3 |
+
import cv2
|
4 |
+
import numpy as np
|
5 |
+
from skimage.io import imread
|
6 |
+
from skimage.transform import resize
|
7 |
+
from skimage.feature import hog
|
8 |
+
import torch
|
9 |
+
import torch.nn as nn
|
10 |
+
import torchvision.models as models
|
11 |
+
import torchvision.transforms as transforms
|
12 |
+
from PIL import Image
|
13 |
+
import pickle
|
14 |
+
|
15 |
+
# Load the pretrained resnet-50 model
|
16 |
+
resnet = models.resnet50(pretrained=True)
|
17 |
+
resnet = nn.Sequential(*list(resnet.children())[:-1])
|
18 |
+
resnet.eval()
|
19 |
+
|
20 |
+
# Load the demo model
|
21 |
+
with open('demo.pkl', 'rb') as f:
|
22 |
+
demo = pickle.load(f)
|
23 |
+
|
24 |
+
# defining some of the important functions that are needed
|
25 |
+
|
26 |
+
# defining feature extraction for resnet-50
|
27 |
+
def extract_features(image, model):
|
28 |
+
image_np = np.array(image)
|
29 |
+
preprocess = transforms.Compose([
|
30 |
+
transforms.ToPILImage(),
|
31 |
+
transforms.Resize(256),
|
32 |
+
transforms.CenterCrop(224),
|
33 |
+
transforms.ToTensor(),
|
34 |
+
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
|
35 |
+
])
|
36 |
+
image = preprocess(image_np)
|
37 |
+
image = image.unsqueeze(0)
|
38 |
+
with torch.no_grad():
|
39 |
+
features = model(image)
|
40 |
+
features = features.squeeze(0)
|
41 |
+
return features
|
42 |
+
|
43 |
+
|
44 |
+
# hog calculation
|
45 |
+
def compute_hog(img):
|
46 |
+
resized_img = resize(img, (128*4, 64*4))
|
47 |
+
fd, hog_image = hog(resized_img, orientations=9, pixels_per_cell=(8, 8),
|
48 |
+
cells_per_block=(2, 2), visualize=True,channel_axis=-1)
|
49 |
+
return fd
|
50 |
+
|
51 |
+
def get_pixel(img, center, x, y):
|
52 |
+
new_value = 0
|
53 |
+
try:
|
54 |
+
if img[x][y] >= center:
|
55 |
+
new_value = 1
|
56 |
+
except:
|
57 |
+
pass
|
58 |
+
return new_value
|
59 |
+
|
60 |
+
# calculate lbp
|
61 |
+
|
62 |
+
def lbp_calculated_pixel(img, x, y):
|
63 |
+
center = img[x][y]
|
64 |
+
val_ar = []
|
65 |
+
val_ar.append(get_pixel(img, center, x-1, y+1))
|
66 |
+
val_ar.append(get_pixel(img, center, x, y+1))
|
67 |
+
val_ar.append(get_pixel(img, center, x+1, y+1))
|
68 |
+
val_ar.append(get_pixel(img, center, x+1, y))
|
69 |
+
val_ar.append(get_pixel(img, center, x+1, y-1))
|
70 |
+
val_ar.append(get_pixel(img, center, x, y-1))
|
71 |
+
val_ar.append(get_pixel(img, center, x-1, y-1))
|
72 |
+
val_ar.append(get_pixel(img, center, x-1, y))
|
73 |
+
|
74 |
+
power_val = [1, 2, 4, 8, 16, 32, 64, 128]
|
75 |
+
val = 0
|
76 |
+
for i in range(len(val_ar)):
|
77 |
+
val += val_ar[i] * power_val[i]
|
78 |
+
return val
|
79 |
+
|
80 |
+
def calcLBP(img):
|
81 |
+
height, width, channel = img.shape
|
82 |
+
img_gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
|
83 |
+
img_lbp = np.zeros((height, width,3), np.uint8)
|
84 |
+
for i in range(0, height):
|
85 |
+
for j in range(0, width):
|
86 |
+
img_lbp[i, j] = lbp_calculated_pixel(img_gray, i, j)
|
87 |
+
hist_lbp = cv2.calcHist([img_lbp], [0], None, [256], [0, 256])
|
88 |
+
return hist_lbp.flatten()
|
89 |
+
|
90 |
+
# Function to infer the class of an image
|
91 |
+
|
92 |
+
def infer(image_path):
|
93 |
+
image = imread(image_path) # reading the image
|
94 |
+
hog_feature = compute_hog(image) # hog features
|
95 |
+
lbp_feature = calcLBP(image) # lbp features
|
96 |
+
cnn_feature = extract_features(image, resnet).numpy() #cnn features
|
97 |
+
hog_feature = hog_feature.reshape(-1) # reshaping
|
98 |
+
lbp_feature = lbp_feature.reshape(-1) #reshaping
|
99 |
+
cnn_feature = cnn_feature.flatten() # cnn features
|
100 |
+
combined_feature = np.concatenate((hog_feature, lbp_feature, cnn_feature)) # combining all the features
|
101 |
+
prediction = demo.predict([combined_feature])
|
102 |
+
return prediction[0]
|
103 |
+
# Streamlit code
|
104 |
+
|
105 |
+
st.title("Face Identification")
|
106 |
+
uploaded_file = st.file_uploader("Choose an image...", type=["png"])
|
107 |
+
|
108 |
+
if uploaded_file is not None:
|
109 |
+
with open("temp_image.png", "wb") as f:
|
110 |
+
f.write(uploaded_file.getvalue())
|
111 |
+
image = Image.open("temp_image.png")
|
112 |
+
st.image(image, caption='Uploaded Image', use_column_width=True)
|
113 |
+
st.write("")
|
114 |
+
st.write("Classifying...")
|
115 |
+
prediction = infer("temp_image.png")
|
116 |
+
st.write("Prediction:", prediction)
|
demo.pkl
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:332ac3f12fcbf253869c688b99862faec19190e447465b080eaf5a5e049ecd57
|
3 |
+
size 4067527
|
requirements.txt
ADDED
@@ -0,0 +1,10 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
opencv-python
|
2 |
+
numpy
|
3 |
+
scikit-image
|
4 |
+
torch
|
5 |
+
torchvision
|
6 |
+
streamlit
|
7 |
+
Pillow
|
8 |
+
pickle-mixin
|
9 |
+
scikit-learn
|
10 |
+
|