import io
import json
import torch
from PIL import Image
from transformers import DetrImageProcessor, DetrForObjectDetection

# Initialize the DETR model and processor
processor = DetrImageProcessor.from_pretrained("facebook/detr-resnet-50", revision="no_timm")
model = DetrForObjectDetection.from_pretrained("facebook/detr-resnet-50", revision="no_timm")

def predict(inputs):
    # Load the image from the provided inputs
    image = Image.open(io.BytesIO(inputs["image"]))

    # Preprocess the image
    inputs = processor(images=image, return_tensors="pt")

    # Perform object detection
    outputs = model(**inputs)

    # Convert outputs to COCO API format
    target_sizes = torch.tensor([image.size[::-1]])
    results = processor.post_process_object_detection(outputs, target_sizes=target_sizes, threshold=0.9)[0]

    # Prepare the results in a dictionary format
    detections = [{"label": model.config.id2label[label.item()], "confidence": score.item(), "box": box.tolist()} 
                  for score, label, box in zip(results["scores"], results["labels"], results["boxes"])]

    return detections

# Define the API endpoint for Hugging Face Spaces
def huggingface_spaces_endpoint(inputs):
    # Call the predict function with the provided inputs
    detections = predict(inputs)

    # Return the detections as a JSON object
    return json.dumps(detections)