from transformers import DetrImageProcessor, DetrForObjectDetection
from PIL import Image, ImageDraw
import torch
import gradio as gr
import requests
from io import BytesIO

# Load pre-trained DETR model
processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50")
model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50")

# COCO class index for "person" = 1 (used as proxy for face detection)
FACE_CLASS_INDEX = 1

def detect_faces(img: Image.Image):
    # Prepare input for the model
    inputs = processor(images=img, return_tensors="pt")
    outputs = model(**inputs)

    # Get outputs
    target_sizes = torch.tensor([img.size[::-1]])
    results = processor.post_process_object_detection(outputs, target_sizes=target_sizes, threshold=0.9)[0]

    # Draw bounding boxes
    draw = ImageDraw.Draw(img)
    for score, label, box in zip(results["scores"], results["labels"], results["boxes"]):
        if label.item() == FACE_CLASS_INDEX:  # 'person'
            box = [round(i, 2) for i in box.tolist()]
            draw.rectangle(box, outline="green", width=3)
            draw.text((box[0], box[1]), f"{score:.2f}", fill="green")

    return img

# Gradio interface
iface = gr.Interface(
    fn=detect_faces,
    inputs=gr.Image(type="pil"),
    outputs="image",
    title="Face Detection App (Hugging Face + Gradio)",
    description="Upload an image and detect faces using facebook/detr-resnet-50 model."
)

iface.launch()