hah1504 / Find_the_right_glass_frame

Glass Frame Preview

Table of Contents

• Project Overview
• Algorithmic Steps
• Usage
• Code
• Future Works
• References

Deployed App

Presentation

---

Back to Table of Content

Project Overview

When someone wants to get a new glass frame, he/she has to go through a process of trial and error. Maybe he/she finds some eyeglasses attractive on display in the optical shop, but not so great when he/she try them on. Depending on face shapes, skin tones, hair color and eye color, a certain glass frame looks awesome to a certain person. So, sometimes it become quite hard for a person to choose a right glass frame in a optical shop.
Artificial Intelligence (AI) can come as rescue in this scenario. AI is already changing almost every spheres of our lives. With the recent advancement of deep learning technology and easily available cloud and edge computing power, we can leverage AI to help us choose the right glass frame for us. This web app will show a basic usage of AI with Intel OpenVINO toolkit to help a person to choose a right glass frame virtually without all the hassles and dilemma.

---

Back to Table of Content

Algorithmic Steps

Step 1: Input Image Preprocessing

Step 2: Facial Landmarks Estimation

Step 3: Overlay Transparent Filter

Refer to the Code for more details.

---

Back to Table of Content

Usage

Initialize OpenVINO environment

Facial Landmarks model

python app.py -i "images/person1.jpg" -t "FACIAL" -m "models/facial-landmarks-35-adas-0002.xml"

Overlay Glass Filter

python app.py -i "images/person1.jpg" -t "GLASS" -m "models/facial-landmarks-35-adas-0002.xml" -g "images/glasses/glass9.png"

Age and Gender Model

python app.py -i "images/person1.jpg" -t "GENDER" -m "models/age-gender-recognition-retail-0013.xml"

---

Back to Table of Content

Code

Handle Facial Landmarks Metadata Model

def handle_facial(output, input_shape):
    '''
    Handles the output of the Facial landmarks Metadata model.
    Returns the coordinates of 35 facial landmarks.
    '''
    #print(output.keys())
    fc=output['align_fc3']
    
    coords=[]
    for i in range(0,len(fc[0]),2):
        x=int(fc[0][i]*input_shape[1])
        y=int(fc[0][i+1]*input_shape[0])
        coords.append(x)
        coords.append(y)
    return coords

Handle Age and Gender Model

def handle_gender(output, input_shape):
    '''
    Handles the output of the Gender and Age Metadata model.
    Returns the age and gender: argmax of softmax output
    '''
    #print(output.keys())
    age=int(output["age_conv3"].flatten()[0]*100)
    gender_type=output["prob"].flatten()
    gender_class=np.argmax(gender_type) 

    return age, gender_class

Output Facial Landmarks Model Image

    if model_type == "FACIAL":
        for i in range(0,len(output),2):
            cv2.circle(image, (output[i],output[i+1]), 4, (255, 0, 0), -1)

            font                   = cv2.FONT_HERSHEY_SIMPLEX
            bottomLeftCornerOfText = (output[i],output[i+1])
            fontScale              = 0.7
            fontColor              = (255,255,255)
            lineType               = 2

            cv2.putText(image,'p'+str(int(i/2)), 
                bottomLeftCornerOfText, 
                font, 
                fontScale,
                fontColor,
                lineType)             
        return image

Ouput Glass Filter Model Image

    if model_type == "GLASS":
        image_copy = np.copy(image)
        glasses=cv2.imread(glass,-1)
        scale=((output[36]-output[68])**2+(output[37]-output[69])**2)**(1/2.0)
        glasses=cv2.resize(glasses,(int(scale),int(scale*glasses.shape[0]/glasses.shape[1])))
        translation_vertical=int((output[1]+output[5])/2-glasses.shape[1]/2)
        translation_horizontal=int((output[0]+output[4])/2-glasses.shape[0]/2)
        gw,gh,gc = glasses.shape
        for i in range(0,gw):       # Overlay the filter based on the alpha channel(glass)
            for j in range(0,gh):
                if glasses[i,j][3] != 0:
                    image[i+translation_vertical,j+translation_horizontal]=glasses[i,j][:-1]
        return image

Output Age and Gender Model Image

    if model_type == "GENDER":
        #print(output[0], output[1])
        age = output[0]
        #print(age)
        gender_type = GENDER_TYPES[output[1]]
        #print(gender_type)
        scaler = max(int(image.shape[0] / 5000), 1)
        image = cv2.putText(image, 
            "{},{} ".format(age,gender_type), 
            (20, image.shape[0]-10), cv2.FONT_HERSHEY_SIMPLEX, 
            2 * scaler, (0,255, 0), 3 * scaler)
        return image

---

Back to Table of Content

Future Works

---

Back to Table of Content

References

1. URL of Deployed App
2. Presentation
3. The best glasses for your face shape and skin tone
4. Host a Static Website

---

Back to Table of Content