app.py

import gradio as gr
import cv2
import onnxruntime
import numpy as np

onnx_model_vgg19_path = "./vgg19-30epochs.onnx"
onnx_model_inceptionv3_path = "./InceptionV3-20epochs.onnx"
onnx_model_resnet101_path = "./Resnet101-30epochs.onnx"
onnx_model_vgg16_path = "./vgg16-20epochs.onnx"

labels = ['Ajloun Castle', 'Hadrians Arch', 'Petra-siq', 'Roman Ruins-Jerash', 'Roman amphitheater', 'The Cardo Maximus of Jerash', 'Wadi Rum', 'petra-Treasury', 'umm qais']

def predict_image(image_path, model):
    if model == "InceptionV3":
        img_size = (550, 475)
        model_inceptionv3 = onnxruntime.InferenceSession(onnx_model_inceptionv3_path)
    elif model == "Resnet101":
        img_size = (250, 200)
        model_resnet101 = onnxruntime.InferenceSession(onnx_model_resnet101_path)
    elif model == "Vgg19":
        img_size = (250, 200)
        model_vgg19 = onnxruntime.InferenceSession(onnx_model_vgg19_path)
    elif model == "Vgg16":
        img_size = (200, 150)
        model_vgg16 = onnxruntime.InferenceSession(onnx_model_vgg16_path)

    img = cv2.imread(image_path)
    img = cv2.resize(img, img_size)
    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
    img = np.expand_dims(img, axis=0)  # Add batch dimension
    img = img.astype(np.float32) / 255.0  # Normalize pixel values

    if model == "Vgg19":
        input_name = model_vgg19.get_inputs()[0].name
        output_name = model_vgg19.get_outputs()[0].name
        prediction = model_vgg19.run(None, {input_name: img})
    elif model == "Vgg16":
        input_name = model_vgg16.get_inputs()[0].name
        output_name = model_vgg16.get_outputs()[0].name
        prediction = model_vgg16.run(None, {input_name: img})
    elif model == "InceptionV3":
        input_name = model_inceptionv3.get_inputs()[0].name
        output_name = model_inceptionv3.get_outputs()[0].name
        prediction = model_inceptionv3.run(None, {input_name: img})
    elif model == "Resnet101":
        input_name = model_resnet101.get_inputs()[0].name
        output_name = model_resnet101.get_outputs()[0].name
        prediction = model_resnet101.run(None, {input_name: img})

    softmax_output = np.exp(prediction[0][0]) / np.sum(np.exp(prediction[0][0]), axis=-1)
    max_confidence = np.max(softmax_output)
    predicted_class = np.argmax(softmax_output)

    if max_confidence < 0.2:
        return "Hmm, it's a bit tricky. Feel free to add another image, and I'll do my best to make a guess!"

    predicted_label = labels[predicted_class]
    return predicted_label

inputs_image = [
    gr.Image(type="filepath", label="Input Image"),
    gr.Radio(choices=[
        "Vgg16",
        "InceptionV3",
        "Vgg19",
        "Resnet101",
    ], type="value", label="Select_model")
]
outputs_text = [gr.Textbox(label="Site")]
interface_image = gr.Interface(
    inputs=inputs_image,
    fn=predict_image,
    outputs=outputs_text,
    title="classifier_demo"
)
interface_image.launch()