app.py

import gradio as gr
import os
import torch
from torch import nn
import torchvision
from timeit import default_timer as timer
from typing import Tuple, Dict
from PIL import Image  # Added for image verification and conversion

def create_effnetb2_model(num_classes: int = 3, seed: int = 42):
    """Creates an EfficientNetB2 feature extractor model and transforms."""
    weights = torchvision.models.EfficientNet_B2_Weights.DEFAULT
    transforms = weights.transforms()
    model = torchvision.models.efficientnet_b2(weights=weights)

    for param in model.parameters():
        param.requires_grad = False

    torch.manual_seed(seed)
    model.classifier = nn.Sequential(
        nn.Dropout(p=0.3, inplace=True),
        nn.Linear(in_features=1408, out_features=num_classes),
    )
    
    return model, transforms


# Load class names
with open("class_names.txt", "r") as f:
    class_names = [food_name.strip() for food_name in f.readlines()]

# Create model and transforms
effnetb2, effnetb2_transforms = create_effnetb2_model(num_classes=101)

# Load pretrained weights
effnetb2.load_state_dict(
    torch.load(
        f="pretrained_effnetb2_feature_extractor_food101_20_percent.pth",
        map_location=torch.device("cpu"),
    )
)


def predict(img) -> Tuple[Dict, float]:
    """Transforms and performs a prediction on img and returns prediction and time taken."""
    start_time = timer()
    
    # Convert to RGB to avoid dtype issues
    if img.mode != "RGB":
        img = img.convert("RGB")

    # Apply transforms
    img = effnetb2_transforms(img).unsqueeze(0)

    # Inference
    effnetb2.eval()
    with torch.inference_mode():
        pred_probs = torch.softmax(effnetb2(img), dim=1)

    pred_labels_and_probs = {
        class_names[i]: float(pred_probs[0][i]) for i in range(len(class_names))
    }
    pred_time = round(timer() - start_time, 5)
    return pred_labels_and_probs, pred_time


# Verify examples directory and images
example_list = []
if os.path.exists("examples"):
    for example in os.listdir("examples"):
        example_path = os.path.join("examples", example)
        try:
            img = Image.open(example_path)
            img.verify()  # Verify image is not corrupted
            example_list.append([example_path])
        except Exception as e:
            print(f"Skipping example {example}: {e}")

# Gradio Interface Setup
title = "FoodVision Big 🍔👁"
description = "An EfficientNetB2 feature extractor computer vision model to classify images of food into 101 different classes."
article = "Created by [Ali Khalaji](https://github.com/codali-ml)."

demo = gr.Interface(
    fn=predict,
    inputs=gr.Image(type="pil"),
    outputs=[
        gr.Label(num_top_classes=5, label="Predictions"),
        gr.Number(label="Prediction time (s)"),
    ],
    examples=example_list,
    title=title,
    description=description,
    article=article,
)

demo.launch()