app.py

import gradio as gr
from create_effnet import create_effnetb2_instance
from typing import Dict, Tuple, List
from PIL import Image
from torch import nn
import torchvision
from pathlib import Path
from timeit import default_timer as timer
import torch

def predict(img,
            classes:List[str]=['pizza', 'steak', 'sushi']
            )->Tuple[Dict, int]:
  # Device
  model, transform = create_effnetb2_instance(num_classes=len(classes),
                                              device='cpu')
  
  weight_path = Path("EfficientNet_B2_10_Epochs.pth")
  
  # Forgot to load the weights
  model.load_state_dict(state_dict=torch.load(f=weight_path,
                                              map_location="cpu"))

  # Start timer
  start = timer()

  # Transform the Image
  transformed_image = transform(img)

  # Eval mode and get prediction results
  model.eval()
  with torch.inference_mode():
    logit = model(transformed_image.unsqueeze(dim=0))
    pred_probs = torch.softmax(input=logit,
                               dim=1)
    preds = torch.argmax(input=pred_probs,
                         dim=1).item()

  # Stop the Timer
  end = timer()
  # Create Dictionary for Prediction Probabilities
  pred_label_prob_dict = dict()
  for idx, probs in enumerate(pred_probs.squeeze(dim=0).tolist()):
    class_name = classes[idx]
    pred_label_prob_dict[class_name] = probs

  # Calculate prediction time
  pred_time = round(end-start, 4)

  return (pred_label_prob_dict, pred_time)

example_img_path = Path("examples")
title = "Pizza 🍕, Steak 🥩, Sushi 🍣 Classifier"
description = "Image Classifier which works on Feature Extractor EffNet B2 model and trained on Pizza, Steak, Sushi dataset"
example_images = [[img] for img in example_img_path.iterdir()]

interface = gr.Interface(
    fn=predict,
    inputs=gr.Image(type="pil"),
    outputs=[gr.Label(num_top_classes=3,
                     label="Model thinks"),
             gr.Number(label="Time Taken (in seconds)")],
    title=title,
    description=description,
    examples=example_images)


if __name__ == "__main__":
  interface.launch(share=True)