Proper way to convert an image to TF record Format (Writing an image to TFrecord format generate Sparse Tensor)

I am reading an image from the local file system, converting it to bytes format and finally ingesting the image to tf.train.Feature to convert into TFRecord format. Things are working fine until the moment I read the TFrecord and extract the image bytes format which seems to be a sparse format in the end. Below is my code for the complete process flow.

reading df and image file: No Error

import tensorflow as tf
from PIL import Image
img_bytes_list = []
for img_path in df.filepath:
    with tf.io.gfile.GFile(img_path, "rb") as f:
        raw_img = f.read()
        img_bytes_list.append(raw_img)

defining features : No Error

write_features = {'filename': tf.train.Feature(bytes_list=tf.train.BytesList(value=df['filename'].apply(lambda x: x.encode("utf-8")))),
                    'img_arr':tf.train.Feature(bytes_list=tf.train.BytesList(value=img_bytes_list)),
                    'width': tf.train.Feature(int64_list=tf.train.Int64List(value=df['width'])),
                    'height': tf.train.Feature(int64_list=tf.train.Int64List(value=df['height'])), 
                    'img_class': tf.train.Feature(bytes_list=tf.train.BytesList(value=df['class'].apply(lambda x: x.encode("utf-8")))),
                    'xmin': tf.train.Feature(int64_list=tf.train.Int64List(value=df['xmin'])),
                    'ymin': tf.train.Feature(int64_list=tf.train.Int64List(value=df['ymin'])),
                    'xmax': tf.train.Feature(int64_list=tf.train.Int64List(value=df['xmax'])),
                    'ymax': tf.train.Feature(int64_list=tf.train.Int64List(value=df['ymax']))}

create example: No Error

example = tf.train.Example(features=tf.train.Features(feature=write_features))

writing data in TfRecord Format: No Error

with tf.io.TFRecordWriter('image_data_tfr') as writer:
    writer.write(example.SerializeToString())

Read and print data: No Error

read_features = {"filename": tf.io.VarLenFeature(dtype=tf.string),
                "img_arr": tf.io.VarLenFeature(dtype=tf.string),
                "width": tf.io.VarLenFeature(dtype=tf.int64),
                "height": tf.io.VarLenFeature(dtype=tf.int64),
                "class": tf.io.VarLenFeature(dtype=tf.string),
                "xmin": tf.io.VarLenFeature(dtype=tf.int64),
                "ymin": tf.io.VarLenFeature(dtype=tf.int64),
                "xmax": tf.io.VarLenFeature(dtype=tf.int64),
                "ymax": tf.io.VarLenFeature(dtype=tf.int64)}

reading single example from tfrecords format: No Error

for serialized_example in tf.data.TFRecordDataset(["image_data_tfr"]):
    parsed_s_example = tf.io.parse_single_example(serialized=serialized_example,
                                                features=read_features)

reading image data from tfrecords format: No Error

image_raw = parsed_s_example['img_arr']

encoded_jpg_io = io.BytesIO(image_raw)

Here it is giving error: TypeError: a bytes-like object is required, not 'SparseTensor'

image = Image.open(encoded_jpg_io)
width, height = image.size
print(width, height)

Please tell me what changes are required at the input of "image_arr" so that it will not generate sparse tensor and return a byte format ?

Is there anything that I can do to optimize my existing code?