app.py

import gradio as gr
import torch
import onnxruntime as ort
import os
import gdown
from PIL import Image
import numpy as np

means = np.array([0.485, 0.456, 0.406])
stds = np.array([0.229, 0.224, 0.225])

# Define the model URL and output path
model_url = "https://drive.google.com/file/d/18HYScsRJuRmfzL0E0BW35uaA542Vd5M5/view?usp=sharing"
model_path = os.path.join(os.getcwd(),"bone_age_model.onnx")

# Check if the model file exists and download if it does not
if not os.path.exists(model_path):
    gdown.download(model_url, model_path, fuzzy=True, quiet=False)

# Initialize the ONNX session
session = ort.InferenceSession(model_path)

# Define the inference function
def inference(sample_name):
    sample_path = os.path.join(os.getcwd(),'samples', f'{sample_name}.pth')
    sample = torch.load(sample_path)
    age = sample['boneage'].item()
    outputs = session.run(None, {"input": sample['path'].numpy()})
    predicted_age = (outputs[0]*41.172)+127.329
    # Get the image data from the MetaTensor and convert it to a format PIL can handle
    image_data = sample['path'][0].numpy()
    # Denormalize the image data
    for i in range(3):  # Assuming the image has 3 channels
        image_data[i,:,:] = image_data[i,:,:] * stds[i] + means[i]
    # Rescale to [0, 1]
    # Convert to [0, 255] and to uint8
    image_data = (image_data * 255).astype(np.uint8)
    # Remove any singleton dimensions if necessary
    image_data = np.moveaxis(image_data,0,-1)
    image = Image.fromarray(image_data)
    return age, round(predicted_age[0][0]), image

# List of sample file names
sample_files = sorted(os.listdir(os.path.join(os.getcwd(),'samples')))
sample_names = [os.path.basename(x).split('.pth')[0] for x in sample_files]
# Create Gradio interface
dropdown = gr.Dropdown(choices=sample_names, label="Select a sample")

iface = gr.Interface(
    fn=inference,
    inputs=dropdown,
    outputs=[
        gr.Textbox(label="Bone Age"), 
        gr.Textbox(label="Predicted Bone Age"),
        gr.Image(label="Image")
    ],
    title="Bone Age Prediction",
    description="Select a sample from the dropdown to see the bone age and predicted bone age."
)

# Launch the app
iface.launch()