📄

4) Object Detection

Object Detection using Yolo

Checking and Changing the version of Keras

•

Check the version: pip show keras

•

Change the version: pip install keras==[target version]

pip show keras
Python
복사

pip install keras==2.6.0
Python
복사

Yolo3 weights Download

•

Before run, please download yolo-tiny.weights file

•

Or, Github Repository clone

Convert the format of Yolo3 file

•

Convert yolov3 file to .h5 format, which can be used in Keras

•

Using Keras-yolo3’s convert.py code

◦

Don’t need to download if you clonde the github repository above.

•

File description

◦

convert.py : run to convert the file format

◦

yolov3.cfg : definition of model structure which is used in Darknet

◦

yolov3.weight : weight of trained model from Darknet

!python convert.py yolov3-tiny.cfg yolov3-tiny.weights model_data/yolo_tiny.h5
Python
복사

[Error]

Test with Sample Image

from IPython.display import display
from PIL import Image
from yolo import YOLO

import tensorflow.compat.v1.keras.backend as K
import tensorflow as tf
tf.compat.v1.disable_eager_execution()

def objectDetection(file, model_path, class_path):
    yolo = YOLO(model_path=model_path, classes_path=class_path, anchors_path="model_data/yolo_tiny_anchors.txt")
    image = Image.open(file)
    result_image = yolo.detect_image(image)
    display(result_image)

objectDetection('dog.jpg', 'model_data/yolo_tiny.h5', 'model_data/coco_classes.txt')
Python
복사

[Error]

Result

Test with Real Image

from IPython.display import display
from PIL import Image
from yolo import YOLO

import tensorflow.compat.v1.keras.backend as K
import tensorflow as tf
tf.compat.v1.disable_eager_execution()

def objectDetection(file, model_path, class_path):
    yolo = YOLO(model_path=model_path, classes_path=class_path, anchors_path="model_data/yolo_tiny_anchors.txt")
    image = Image.open(file)
    result_image = yolo.detect_image(image)
    display(result_image)

objectDetection('resize_img/00000.jpg', 'model_data/yolo_tiny.h5', 'model_data/coco_classes.txt')
Python
복사

Result

Image annotation

•

Converting the labelling data to train the model

import xml.etree.ElementTree as ET
from os import getcwd
import glob

classes = ['beam']

def convert_annotation(annotation_voc, train_all_file):
    tree = ET.parse(annotation_voc)
    objects = tree.findall('./object')

    for obj in objects:
        difficult = obj.find('difficult').text
        cls = obj.find('name').text
        if cls not in classes or int(difficult)==1: continue
        cls_id = classes.index(cls)
        xmlbox = obj.find('bndbox')
        b = (int(float(xmlbox.find('xmin').text)), int(float(xmlbox.find('ymin').text)), int(float(xmlbox.find('xmax').text)), int(float(xmlbox.find('ymax').text)))
        train_all_file.write(" " + ",".join([str(a) for a in b]) + ',' + str(cls_id))

train_all_file = open('./data/personal/annotations_yolo/train_all.txt', 'w')

# Get annotations_voc list
for xml in xml_list:
    annotations_voc = glob.glob(f'./data/personal/annotations_voc/{xml}')
    for annotation_voc in annotations_voc:
        image_id = annotation_voc.split('/')[-1].split('.')[0]+'.JPG'
        print(image_id)
        train_all_file.write(f'./data/personal/image_train/{image_id}')
        convert_annotation(annotation_voc, train_all_file)
        train_all_file.write('\n')
train_all_file.close()
Python
복사

Result