Best Practices for Loading Images in TensorFlow Object Detection Pipeline

[silverhawkjr11]

I'm currently working on an object detection project using TensorFlow, and I'm seeking advice on the best practices for loading images into a dataset
im trying to implement this but with a custom dataset:
https://colab.research.google.com/github/keras-team/keras-io/blob/master/guides/ipynb/keras_cv/object_detection_keras_cv.ipynb
Below is an example of my current dataset file architecture:

my_dataset/
|-- train/
|   |-- images/
|   |   |-- image1.jpg
|   |   |-- image2.jpg
|   |   |-- ...
|   |-- annotations/
|   |   |-- image1.xml
|   |   |-- image2.xml
|   |   |-- ...
|-- valid/
|   |-- images/
|   |   |-- image3.jpg
|   |   |-- image4.jpg
|   |   |-- ...
|   |-- annotations/
|   |   |-- image3.xml
|   |   |-- image4.xml
|   |   |-- ...
|-- test/
|   |-- images/
|   |   |-- image5.jpg
|   |   |-- image6.jpg
|   |   |-- ...
|   |-- annotations/
|   |   |-- image5.xml
|   |   |-- image6.xml
|   |   |-- ...

heres an example of the annotations:

<annotation>
	<folder>VOC2007</folder>
	<filename>000002.jpg</filename>
	<source>
		<database>The VOC2007 Database</database>
		<annotation>PASCAL VOC2007</annotation>
		<image>flickr</image>
		<flickrid>329145082</flickrid>
	</source>
	<owner>
		<flickrid>hiromori2</flickrid>
		<name>Hiroyuki Mori</name>
	</owner>
	<size>
		<width>335</width>
		<height>500</height>
		<depth>3</depth>
	</size>
	<segmented>0</segmented>
	<object>
		<name>train</name>
		<pose>Unspecified</pose>
		<truncated>0</truncated>
		<difficult>0</difficult>
		<bndbox>
			<xmin>139</xmin>
			<ymin>200</ymin>
			<xmax>207</xmax>
			<ymax>301</ymax>
		</bndbox>
	</object>
</annotation>

thats my approach which is not working properly:

def parse_annotation(annotation_path, bounding_box_format):
    tree = ET.parse(annotation_path)
    root = tree.getroot()

    boxes = []
    for obj in root.findall('object'):
        name = obj.find('name').text
        box = obj.find('bndbox')
        xmin = int(box.find('xmin').text)
        ymin = int(box.find('ymin').text)
        xmax = int(box.find('xmax').text)
        ymax = int(box.find('ymax').text)

        if bounding_box_format == 'xywh':
            x = (xmin + xmax) / 2.0
            y = (ymin + ymax) / 2.0
            w = xmax - xmin
            h = ymax - ymin
            boxes.append({"class": name, "x": x, "y": y, "w": w, "h": h})
        elif bounding_box_format == 'yxyx':
            boxes.append({"class": name, "ymin": ymin, "xmin": xmin, "ymax": ymax, "xmax": xmax})

    return boxes
def load_custom_dataset(split, path, bounding_box_format):
    images_path = os.path.join(path, split, 'images')
    annotations_path = os.path.join(path, split, 'annotations')
    dataset = []

    for filename in os.listdir(annotations_path):
        if filename.endswith('.xml'):
            annotation_path = os.path.join(annotations_path, filename)
            image_filename = filename.replace('.xml', '.jpg')
            image_path = os.path.join(images_path, image_filename)

            if os.path.exists(image_path):
                image = tf.image.decode_jpeg(tf.io.read_file(image_path), channels=3)
                # Further processing...
                bounding_boxes = parse_annotation(annotation_path, bounding_box_format)
                
                if bounding_boxes:
                    dataset.append({"image": image, "bounding_boxes": bounding_boxes})

    return tf.data.Dataset.from_tensor_slices(dataset)
    
train_ds = load_custom_dataset(split="train", path="ds", bounding_box_format="xywh")

my overall goal is to replace the tfds.load in their example with my own dataset