ViT-MRI-FineTuning / preprocessing.py
Tanusree88's picture
Rename app.py to preprocessing.py
3c12f92 verified
raw
history blame
1.61 kB
import os
import numpy as np
import torch
from transformers import ViTForImageClassification, ViTImageProcessor
import nibabel as nib # For loading .nii files
from PIL import Image # For loading .jpg and .jpeg files
# Function to preprocess images based on their file format
def preprocess_image(image_path):
ext = os.path.splitext(image_path)[-1].lower() # Get the file extension
# Case 1: .nii files (NIfTI format)
if ext == '.nii' or ext == '.nii.gz':
# Load the .nii image
nii_image = nib.load(image_path)
image_data = nii_image.get_fdata()
# Convert to tensor and reshape to [C, H, W] format
image_tensor = torch.tensor(image_data).float()
# Handle cases where the image might have a different shape (e.g., single channel vs multiple channels)
if len(image_tensor.shape) == 3:
image_tensor = image_tensor.unsqueeze(0) # Add channel dimension if not present
# Case 2: .jpg and .jpeg files (JPEG format)
elif ext in ['.jpg', '.jpeg']:
# Load the image using PIL
img = Image.open(image_path).convert('RGB') # Convert to RGB
img = img.resize((224, 224)) # Resize to the input size expected by ViT (224x224)
# Convert to numpy array and then to tensor
img_np = np.array(img)
image_tensor = torch.tensor(img_np).permute(2, 0, 1).float() # Rearrange to [C, H, W]
else:
raise ValueError(f"Unsupported file format: {ext}")
# Normalize image tensor (if required)
image_tensor /= 255.0 # Normalize pixel values to [0, 1]
return image_tensor