app.py

import gradio as gr
import spaces
import torch

from ultralytics import YOLO
from PIL import Image
import supervision as sv
import numpy as np

@spaces.GPU
def yolov8_inference(
    image,
    image_size,
    conf_threshold,
    iou_threshold,
):
    """
    YOLOv8 inference function
    Args:
        image: Input image
        model_path: Path to the model
        image_size: Image size
        conf_threshold: Confidence threshold
        iou_threshold: IOU threshold
    Returns:
        Rendered image
    """
    model = YOLO('erax_nsfw_v1.onnx')
    # set model parameters
    model.overrides['conf'] = conf_threshold  # NMS confidence threshold
    model.overrides['iou'] = iou_threshold  # NMS IoU threshold
    model.overrides['agnostic_nms'] = False  # NMS class-agnostic
    model.overrides['max_det'] = 1000  # maximum number of detections per image

    results = model([image])
    for result in results:
        annotated_image = result.orig_img.copy()
        h, w = annotated_image.shape[:2]
        anchor = h if h > w else w
        
        # make_love class will cover entire context !!!
        # selected_classes = [0, 1, 2, 3, 4, 5] # all classes
        selected_classes = [0, 2, 3, 4, 5] # hidden make_love class
        detections = sv.Detections.from_ultralytics(result)
        detections = detections[np.isin(detections.class_id, selected_classes)]
           
        # box_annotator = sv.BoxAnnotator()
        # annotated_image = box_annotator.annotate(
        #     annotated_image,
        #     detections=detections
        # )
        
        # blur_annotator = sv.BlurAnnotator(kernel_size=anchor/50)    
        # annotated_image = blur_annotator.annotate(
        #     annotated_image.copy(),
        #     detections=detections
        # )
        
        label_annotator = sv.LabelAnnotator(text_color=sv.Color.BLACK,
                                            text_position=sv.Position.CENTER,
                                            text_scale=anchor/1700)
        annotated_image = label_annotator.annotate(
            annotated_image,
            detections=detections
        )
        
        pixelate_annotator = sv.PixelateAnnotator(pixel_size=anchor/50)
        annotated_image = pixelate_annotator.annotate(
            scene=annotated_image.copy(),
            detections=detections
        )
        
        # sv.plot_image(annotated_image, size=(10, 10))
    # results = model.predict(image, imgsz=image_size)
    # render = render_result(model=model, image=image, result=results[0])
    
    return annotated_image[:, :, ::-1]


inputs = [
    gr.Image(type="filepath", label="Input Image"),
    gr.Slider(minimum=320, maximum=1280, value=640, step=320, label="Image Size"),
    gr.Slider(minimum=0.0, maximum=1.0, value=0.25, step=0.05, label="Confidence Threshold"),
    gr.Slider(minimum=0.0, maximum=1.0, value=0.45, step=0.05, label="IOU Threshold"),
]

outputs = gr.Image(type="filepath", label="Output Image")
title = "State-of-the-Art YOLO Models for Object detection"

# examples = [['demo_01.jpg', 'yolov8n', 640, 0.25, 0.45], ['demo_02.jpg', 'yolov8l', 640, 0.25, 0.45], ['demo_03.jpg', 'yolov8x', 1280, 0.25, 0.45]]
demo_app = gr.Interface(
    fn=yolov8_inference,
    inputs=inputs,
    outputs=outputs,
    title=title,
    # examples=examples,
    # cache_examples=True,
)
demo_app.launch(debug=True)