inference.py

import torch
import torchvision.transforms as transforms
from PIL import Image
import os

# Define the device
device = (
    "cuda"
    if torch.cuda.is_available()
    else "mps"
    if torch.backends.mps.is_available()
    else "cpu"
)

class Params:
    def __init__(self):
        self.batch_size = 512
        self.name = "resnet_50"
        self.workers = 16
        self.lr = 0.1
        self.momentum = 0.9
        self.weight_decay = 1e-4
        self.lr_step_size = 30
        self.lr_gamma = 0.1

    def __repr__(self):
        return str(self.__dict__)
    
    def __eq__(self, other):
        return self.__dict__ == other.__dict__
    
params = Params()

# Path to the saved model checkpoint
checkpoint_path = "checkpoints/resnet_50/checkpoint.pth"

# Load the model architecture
from model import ResNet50  # Assuming resnet.py contains your model definition

num_classes = 1000  # Adjust this to match your dataset
model = ResNet50(num_classes=num_classes).to(device)

# Load the trained model weights
checkpoint = torch.load(checkpoint_path)
model.load_state_dict(checkpoint["model"])

model.eval()

# Define transformations for inference
inference_transforms = transforms.Compose([
    transforms.ToTensor(),
    transforms.Resize(size=256),
    transforms.CenterCrop(224),
    transforms.Normalize(mean=[0.485, 0.485, 0.406], std=[0.229, 0.224, 0.225]),
])

# Load class names from the text file
def load_class_names(file_path):
    with open(file_path, 'r') as f:
        class_names = [line.strip() for line in f]
    return class_names

# Function to make predictions on a single image
def predict(image_path, model, transforms, class_names=None):
    # Load and transform the image
    image = Image.open(image_path).convert("RGB")
    image_tensor = transforms(image).unsqueeze(0).to(device)

    # Forward pass
    with torch.no_grad():
        output = model(image_tensor)
        probabilities = torch.nn.functional.softmax(output[0], dim=0)
        top_prob, top_class = probabilities.topk(5, largest=True, sorted=True)

    # Display the top predictions
    print("Predictions:")
    for i in range(top_prob.size(0)):
        class_name = class_names[top_class[i]] if class_names else f"Class {top_class[i].item()}"
        print(f"{class_name}: {top_prob[i].item() * 100:.2f}%")

    return top_prob, top_class

# Path to the ImageNet classes text file
imagenet_classes_file = "imagenet-classes.txt"  # Replace with the actual path to your text file
class_names = load_class_names(imagenet_classes_file)

# Path to the image for inference
image_path = "dog.png"  # Replace with the actual path to your test image

# Make a prediction
predict(image_path, model, inference_transforms, class_names=class_names)