app.py

import gradio as gr
from PIL import Image,ImageDraw, ImageFont, ImageOps
import sys
import torch
from util import Detection, load_font
import os

if os.environ.get('FACE_MODEL') is not None:
    face_model = os.environ.get('FACE_MODEL')
    age_model = os.environ.get('AGE_MODEL')

    torch.hub.download_url_to_file(face_model, 'face_model.pt')
    torch.hub.download_url_to_file(age_model, 'age_model.pt')

sys.path.append("./")
sys.path.append("./yolov5")

from yolov5.detect import predict, load_yolo_model


# Load Models
model, stride, names, pt, jit, onnx, engine = load_yolo_model("face_model.pt", imgsz=[320,320])
age_model_ts = torch.jit.load("age_model.pt")

text_box_height = 22

roboto_font = load_font(height_px=text_box_height-2)


def run_yolo(img0, with_random_augs):

    img0 = ImageOps.contain(img0, (640,640))
    img0 = ImageOps.exif_transpose(img0)

    draw = ImageDraw.Draw(img0)

    predictions = predict(age_model_ts, model, 
        stride, imgsz=[320, 320], 
        conf_thres=0.5, iou_thres=0.45,
        source=img0,
        with_random_augs = with_random_augs
    )

    detections : list[Detection] = []
    for k, bbox  in enumerate(predictions):
        det = Detection(
            (k+1),
            bbox["xmin"],
            bbox["ymin"],
            bbox["xmax"],
            bbox["ymax"],
            bbox["conf"],
            bbox["class"],
            bbox["class"],
            img0.size
        )

        detections.append(det)
        draw.rectangle(((det.xmin, det.ymin), (det.xmax, det.ymax)), fill=None, outline=(255,255,255))

        text_length = roboto_font.getlength(bbox["class"])

        rect_center = (det.xmin + det.xmax - text_length) // 2
        draw.rectangle(((rect_center, det.ymin), (rect_center + text_length, det.ymin + text_box_height)), fill=(255,255,255))
        draw.text((rect_center, det.ymin), det.class_name, fill=(0,0,0), font=roboto_font)

    return img0


""" img = Image.open("D:\\Download\\IMG_20220803_153335c2.jpg").convert("RGB")
run_yolo(img)
sys.exit(1) """


def main():
    input = gr.Image(type='pil', label="Input Image")
    outputs = gr.Image(type="pil", label="Output Image", interactive=False)
    
    augment_preds = gr.Checkbox(label="Apply random augmentations")

    title = "AgeGuesser"
    description = "Guess the age of a person from a facial image!"
    article = """
<p>A fully automated system based on YOLOv5 and EfficientNet to perform face detection and age estimation in real-time.</p>
<p><b>Links</b></p>
<ul>
<li>
<a href='https://link.springer.com/chapter/10.1007/978-3-030-89131-2_25'>Springer</a>
</li>
<li>
<a href='https://www.researchgate.net/publication/355777953_Real-Time_Age_Estimation_from_Facial_Images_Using_YOLO_and_EfficientNet'>Paper</a> 
</li>
<li>
<a href='https://github.com/ai-hazard/AgeGuesser-train'>Github</a> 
</li>
</ul>
<p>Credits to my dear colleague <a href='https://www.linkedin.com/in/nicola-marvulli-904270136/'>Dott. Nicola Marvulli</a>, we've developed AgeGuesser together as part of two university exams. (Computer Vision + Deep Learning)</p>
<p>Credits to my dear professors and the <a href='https://sites.google.com/site/cilabuniba/'>CILAB</a> research group</p>
<ul>
<li>
<a href='https://sites.google.com/site/cilabuniba/people/giovanna-castellano'>Prof. Giovanna Castellano</a>
</li>
<li>
<a href='https://sites.google.com/view/gennaro-vessio/home-page'>Prof. Gennaro Vessio</a> 
</li>
</ul>
    """

    examples = [[Image.open('images/1.jpg').convert("RGB"), False],[Image.open('images/2.jpg').convert("RGB"), False],[Image.open('images/3.jpg').convert("RGB"), False],[Image.open('images/4.jpg').convert("RGB"), False],[Image.open('images/5.jpg').convert("RGB"), False]]

    gr.Interface(run_yolo, [input, augment_preds], outputs, title=title, description=description, article=article, examples=examples, theme="huggingface").launch(enable_queue=True, ) # share=True


main()