app.py

import gradio as gr
import torch
import torchvision.transforms as transforms
from PIL import Image
import torchvision.models as models
import torch.nn as nn

# 🔹 Device setup
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# 🔹 Load ResNet50 and modify for 2-class output
model = models.resnet50(weights=models.ResNet50_Weights.DEFAULT)
in_features = model.fc.in_features
model.fc = nn.Sequential(
    nn.Linear(in_features, 512),
    nn.ReLU(),
    nn.Dropout(0.4),
    nn.Linear(512, 2)  # 2 classes: NORMAL, PNEUMONIA
)

# 🔹 Load the trained model
model.load_state_dict(torch.load("best_model (2).pth", map_location=device))
model.to(device)
model.eval()

# 🔹 Preprocessing - exactly matching your val_transforms
transform = transforms.Compose([
    transforms.Lambda(lambda img: img.convert("RGB")),  # 🧠 Force RGB
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.5]*3, std=[0.5]*3)
])

# 🔹 Class labels
class_names = ["NORMAL", "PNEUMONIA"]

# 🔹 Inference function
def classify_image(img):
    img = transform(img).unsqueeze(0).to(device)
    with torch.no_grad():
        outputs = model(img)
        probs = torch.nn.functional.softmax(outputs, dim=1)
        return {class_names[i]: float(probs[0][i]) for i in range(len(class_names))}

# 🔹 Gradio Interface
interface = gr.Interface(
    fn=classify_image,
    inputs=gr.Image(type="pil"),
    outputs=gr.Label(num_top_classes=2),
    title="🩺 Pneumonia Classifier",
    description="Upload a chest X-ray image. The model predicts whether it's NORMAL or shows signs of PNEUMONIA."
)

# 🔹 Launch the app
interface.launch()