airad.py

import torch
from torch import nn
from torchvision import models, transforms
from PIL import Image
import cv2
import numpy as np
import gdown

class AIRadModel(nn.Module):
  def __init__(self,num_classes=2):
    super(AIRadModel,self).__init__()
    self.model = models.efficientnet_b0(pretrained=False)
    self.num_features = model.classifier[1].in_features
    self.model.classifier = nn.Sequential(
        nn.Dropout(p=0.2),
        nn.Linear(self.num_features, num_classes)  # Two classes: normal, pneumonia
    )

  def forward(self, x):
        return self.model(x)

class AIRadSimModel(nn.Module):
  def __init__(self, num_classes=2):
    super(AIRadSimModel,self).__init__()
    self.sim_model = models.resnet50(pretrained=False)
    self.sim_model.fc = nn.Linear(self.sim_model.fc.in_features,num_classes)

  def forward(self,x):
    return self.sim_model(x)


def load_model():
  model = AIRadModel(num_classes=2)
  file_id = '1CKkdQ5nKWkz3L-ZdgyrJ5SE-oiFwXnSJ'
  gdrive_url = f"https://drive.google.com/uc?id={file_id}"
  model_checkpoint = 'model_checkpoint.pth'
  gdown.download(gdrive_url, model_checkpoint, quiet=False)
  model.load_state_dict(torch.load(model_checkpoint))
  model.eval()
  return model

def load_sim_model():
  sim_model = AIRadSimModel(num_classes=2)
  sim_file_id = 'cjdDsW5QAIlOneOPLg0uYqTURSr0oOLq'
  sim_gdrive_url = f"https://drive.google.com/uc?id={file_id}"
  sim_model_checkpoint = 'sim_model_checkpoint.pth'
  gdown.download(sim_gdrive_url, sim_model_checkpoint, quiet=False)
  sim_model.load_state_dict(torch.load(sim_model_checkpoint))
  sim_model.eval()
  return sim_model()

model = load_model()
sim_model = load_sim_model()
class_names = {0: 'normal', 1: 'pneumonia'}

preprocess = transforms.Compose([
    transforms.Resize(256),
    transforms.CenterCrop(224),
    transforms.ToTensor(),
    transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
])

def predict(image_path):
  image = Image.open(image_path).convert("RGB")
  image_np = np.array(image)
  image_np = cv2.bilateralFilter(image_np, 9, 75, 75)
  image = Image.fromarray(image_np)
  image_tensor = preprocess(image).unsqueeze(0).to(device)

  # Use ResNet50 to predict if the image is an X-ray
  with torch.no_grad():
    sim_output = sim_model(image_tensor)
    _, predicted_sim = torch.max(sim_output, 1)
    predicted_class_sim = predicted_sim.item()

  if predicted_class_sim == 1:  
    with torch.no_grad():
      output = model(image_tensor)
      _, predicted = torch.max(output, 1)
      predicted_class = predicted.item()
      confidence = torch.nn.functional.softmax(output, dim=1)[0][predicted_class].item()
      return class_names[predicted_class] ,confidence

  else:
    return "error"