app.py

import gradio as gr
import torch
from torchvision import transforms
from torchvision.models import resnet18, ResNet18_Weights
from torch import nn
from PIL import Image # pip install pillow

labels = ['fractured','not fractured']

# Same data transformation that was used for inputs (except data augmentation)

imgSize = 128
data_transform = transforms.Compose([
    transforms.Resize(size=(imgSize, imgSize)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406],
                        std=[0.229, 0.224, 0.225])
])

# https://pytorch.org/tutorials/beginner/saving_loading_models.html
# Loading Model for Inference with state_dict (recommended)
model = resnet18(weights=ResNet18_Weights.DEFAULT)
model.fc = nn.Linear(in_features=512, out_features=len(labels))
model.load_state_dict(torch.load("model.pth",map_location=torch.device('cpu')))
model.eval()

def predict(img):
    X = data_transform(img).unsqueeze(0) # returns tensor
    with torch.no_grad():
            predictions = model(X).flatten()
            predictions = torch.nn.functional.softmax(predictions)
            confidences = {labels[i]: float(predictions[i]) for i in range(len(labels))}
    return confidences

title = "Bone Fractures"
description = "Bone fractures classifier trained on the Kaggle dataset using Resnet18"

demo=gr.Interface(fn=predict,
             inputs=gr.Image(type="pil"),
             outputs=gr.Label(num_top_classes=len(labels)),
             title=title,
             description=description,     
             examples=["Elbow_Fractured.jpeg","Finger_Fractured.jpg","Shoulder_Fractured.jpeg","Foot_Not_Fractured.jpg","Leg_Not_Fractured.jpg"])

demo.launch('share=True')