app.py

import gradio as gr

from PIL import Image
import pandas as pd
import numpy as np

import torch
import torch.nn as nn
import torchvision
from torchvision import datasets, models, transforms

from torch_mtcnn import detect_faces
from torch_mtcnn import show_bboxes

def pipeline(img):
    bounding_boxes, landmarks = detect_faces(img)
    if len(bounding_boxes) == 0:
        raise Exception("Didn't find face any faces, try another image!")
    if len(bounding_boxes) > 1:
        raise Exception("Found more than one face, try a profile picture with only one person in it!")
        
    bb = [bounding_boxes[0,0], bounding_boxes[0,1], bounding_boxes[0,2], bounding_boxes[0,3]]
      
    img_cropped = img.crop(bb)
    
    device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
    model_fair_7 = torchvision.models.resnet34(pretrained=True)
    model_fair_7.fc = nn.Linear(model_fair_7.fc.in_features, 18)
    model_fair_7.load_state_dict(torch.load('res34_fair_align_multi_7_20190809.pt', map_location=torch.device('cpu')))
    model_fair_7 = model_fair_7.to(device)
    model_fair_7.eval()

    trans = transforms.Compose([
            transforms.Resize((224, 224)),
            transforms.ToTensor(),
            transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
        ])

    face_names = []
    gender_scores_fair = []
    age_scores_fair = []
    gender_preds_fair = []
    age_preds_fair = []

    image = trans(img_cropped)
    image = image.view(1, 3, 224, 224)  # reshape image to match model dimensions (1 batch size)
    image = image.to(device)

    # fair 7 class
    outputs = model_fair_7(image)
    outputs = outputs.cpu().detach().numpy()
    outputs = np.squeeze(outputs)

    gender_outputs = outputs[7:9]
    age_outputs = outputs[9:18]

    gender_score = np.exp(gender_outputs) / np.sum(np.exp(gender_outputs))
    age_score = np.exp(age_outputs) / np.sum(np.exp(age_outputs))

    gender_pred = np.argmax(gender_score)
    age_pred = np.argmax(age_score)

    gender_scores_fair.append(gender_score)
    age_scores_fair.append(age_score)

    gender_preds_fair.append(gender_pred)
    age_preds_fair.append(age_pred)

    result = pd.DataFrame([gender_preds_fair,
                           age_preds_fair]).T

    result.columns = ['gender_preds_fair',
                      'age_preds_fair']
    # gender
    result.loc[result['gender_preds_fair'] == 0, 'gender'] = 'Male'
    result.loc[result['gender_preds_fair'] == 1, 'gender'] = 'Female'

    # age
    result.loc[result['age_preds_fair'] == 0, 'age'] = '0-2'
    result.loc[result['age_preds_fair'] == 1, 'age'] = '3-9'
    result.loc[result['age_preds_fair'] == 2, 'age'] = '10-19'
    result.loc[result['age_preds_fair'] == 3, 'age'] = '20-29'
    result.loc[result['age_preds_fair'] == 4, 'age'] = '30-39'
    result.loc[result['age_preds_fair'] == 5, 'age'] = '40-49'
    result.loc[result['age_preds_fair'] == 6, 'age'] = '50-59'
    result.loc[result['age_preds_fair'] == 7, 'age'] = '60-69'
    result.loc[result['age_preds_fair'] == 8, 'age'] = '70+'
  
    return "A " + result['gender'][0] + " in the age range of " + result['age'][0]
    
def predict(image):
  try :
      predictions = pipeline(image)
  except Exception as e:
      predictions = e
  return predictions

title = "Estimate age and gender from profile picture."
description = """
This demo is based on the following work:
*Karkkainen, K., & Joo, J. (2021).*
FairFace: Face Attribute Dataset for Balanced Race, Gender, and Age for Bias Measurement and Mitigation. 
In Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (pp. 1548-1558).
See: https://github.com/joojs/fairface
"""
 
gr.Interface(
    predict,
    inputs=gr.inputs.Image(label="Upload a profile picture of a single person", type="pil"),
    outputs=("text"),
    title= title,
    description = description,
    examples=["ex0.jpg","ex4.jpg", "ex1.jpg","ex2.jpg","ex3.jpg","ex5.jpg","ex6.jpg"]
).launch()