app.py

from typing import List
import os
import numpy as np
import torch
import gradio as gr
from PIL import Image
from transformers import AutoImageProcessor, AutoModelForObjectDetection
import supervision as sv
import spaces

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

processor = AutoImageProcessor.from_pretrained("PekingU/rtdetr_r50vd_coco_o365")
model = AutoModelForObjectDetection.from_pretrained("PekingU/rtdetr_r50vd_coco_o365").to(device)

BOUNDING_BOX_ANNOTATOR = sv.BoundingBoxAnnotator()
MASK_ANNOTATOR = sv.MaskAnnotator()
LABEL_ANNOTATOR = sv.LabelAnnotator()
TRACKER = sv.ByteTrack()

def annotate_image(
    input_image,
    detections,
    labels
) -> np.ndarray:
    output_image = MASK_ANNOTATOR.annotate(input_image, detections)
    output_image = BOUNDING_BOX_ANNOTATOR.annotate(output_image, detections)
    output_image = LABEL_ANNOTATOR.annotate(output_image, detections, labels=labels)
    return output_image

def process_image(
    input_image,
    confidence_threshold,
):
    results = query(input_image, confidence_threshold)
    detections = sv.Detections.from_transformers(results[0])
    detections = TRACKER.update_with_detections(detections)
    final_labels = [model.config.id2label[label] for label in detections.class_id.tolist()]
    output_image = annotate_image(input_image, detections, final_labels)
    return output_image

def query(image, confidence_threshold):
    inputs = processor(images=image, return_tensors="pt").to(device)
    with torch.no_grad():
        outputs = model(**inputs)
    target_sizes = torch.tensor([image.size[::-1]])
    results = processor.post_process_object_detection(outputs=outputs, threshold=confidence_threshold, target_sizes=target_sizes)
    return results

def run_demo():
    input_image = gr.inputs.Image(label="Input Image")
    conf = gr.inputs.Slider(label="Confidence Threshold", minimum=0.1, maximum=1.0, value=0.6, step=0.05)
    output_image = gr.outputs.Image(label="Output Image")

    def process_and_display(input_image, conf):
        output_img = process_image(input_image, conf)
        return output_img

    gr.Interface(
        fn=process_and_display,
        inputs=[input_image, conf],
        outputs=output_image,
        title="Real Time Object Detection with RT-DETR",
        description="This Demo uses RT-DETR for object detection in images. Adjust the confidence threshold to see different results.",
        capture_session=True,
    ).launch()