app.py

import gradio as gr
import torch
import torch.nn as nn
from torchvision import transforms, models
from PIL import Image
import numpy as np
import pickle
import os

# Model definition

class FoodIngredientClassifier(nn.Module):
    def __init__(self, num_classes):
        super().__init__()
        self.backbone = models.vit_b_16(weights=None)
        num_features = self.backbone.heads.head.in_features
        self.backbone.heads = nn.Sequential(
            nn.Dropout(0.5),
            nn.Linear(num_features, 1024),
            nn.BatchNorm1d(1024),
            nn.ReLU(),
            nn.Dropout(0.4),
            nn.Linear(1024, 512),
            nn.BatchNorm1d(512),
            nn.ReLU(),
            nn.Dropout(0.3),
            nn.Linear(512, num_classes)
        )

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


# Load model

DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
THRESHOLD = 0.5

def load_model():
    checkpoint = torch.load("model.pth", map_location=DEVICE, weights_only=False)
    mlb = checkpoint["mlb"]
    num_classes = len(mlb.classes_)
    model = FoodIngredientClassifier(num_classes)
    model.load_state_dict(checkpoint["model_state_dict"])
    model.to(DEVICE)
    model.eval()
    return model, mlb

model, mlb = load_model()

# Transform

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

# Inference Function

def predict(image, threshold):
    if image is None:
        return "Please upload an image."

    img = image.convert("RGB")
    img_tensor = transform(img).unsqueeze(0).to(DEVICE)

    with torch.no_grad():
        output = model(img_tensor)
        probs = torch.sigmoid(output).cpu().numpy()[0]

    pred_indices = np.where(probs > threshold)[0]

    if len(pred_indices) == 0:
        return "No ingredients detected above the threshold. Try lowering it."

    results = sorted(
        zip(mlb.classes_[pred_indices], probs[pred_indices]),
        key=lambda x: x[1],
        reverse=True
    )

    output_lines = ["### 🍽️ Detected Ingredients\n"]
    for ingredient, confidence in results:
        bar = "█" * int(confidence * 20)
        output_lines.append(f"**{ingredient}** — {confidence:.1%}  `{bar}`")

    return "\n\n".join(output_lines)


# UI

with gr.Blocks(title="Food Ingredient Detector") as demo:
    gr.Markdown("""
    # Food Ingredient Detector
    Upload a photo of food and the model will identify its ingredients.
    Built with a **ViT-B/16** backbone fine-tuned for multi-label ingredient classification.
    """)

    with gr.Row():
        with gr.Column():
            image_input = gr.Image(type="pil", label="Upload Food Image")
            threshold_slider = gr.Slider(
                minimum=0.1, maximum=0.9, value=0.5, step=0.05,
                label="Detection Threshold",
                info="Lower = more ingredients detected, higher = only confident predictions"
            )
            predict_btn = gr.Button("Detect Ingredients", variant="primary")

        with gr.Column():
            output = gr.Markdown(label="Results")

    predict_btn.click(
        fn=predict,
        inputs=[image_input, threshold_slider],
        outputs=output
    )

    gr.Examples(
        examples=[],  # Add example image paths here if you include sample images
        inputs=image_input
    )

if __name__ == "__main__":
    demo.launch()