Images given to us for manual boxing

zip_file_path = "/tmp/image.zip"

with zipfile.ZipFile(zip_file_path, 'r') as zip_ref:

    zip_ref.extractall("/content/drive/MyDrive/APP/car_plate_box")

Uploads images of rear-end of cars to our G-drive

Coordinates of bounding boxes of our images.

weights_url = "https://storage.googleapis.com/mledu-datasets/inception_v3_weights_tf_dim_ordering_tf_kernels_notop.h5"

weights_file = "inception_v3.h5"

Using pretrained model for our weights as it is more precise.

x = layers.Flatten()(last_output)

x = layers.Dense(512, activation='relu')(x)

final = layers.Dropout(0.2)(x)

a = layers.Dense(1, name="a")(final)

b = layers.Dense(1, name="b")(final)

c = layers.Dense(1, name="c")(final)

d = layers.Dense(1, name="d")(final)

Using LabelImg to annotate bounding boxes as the input to our model so our AI model knows which part of the image needs to be boxed.

The xml file for the bounding boxes we created. Blue highlights the corners of our bounding boxes.

We test our AI model with this image. It successfully creates a bounding box for our Car Plate.

[95, 123, 191, 163]