mosesb/best-comic-panel-detection

1import gradio as gr2from ultralytics import YOLO3import torch4 5model_id = "mosesb/best-comic-panel-detection"6model = YOLO("best.pt")7 8def detect_panels(pil_image, conf_threshold, iou_threshold):9    """10    Takes a PIL image and thresholds, runs YOLOv12 object detection,11    and returns the annotated image with bounding boxes.12    """13    # Run inference on the image with the specified thresholds14    results = model.predict(pil_image, conf=conf_threshold, iou=iou_threshold, verbose=False)15    annotated_image = results[0].plot()16 17    # Gradio's gr.Image component expects an RGB image. The .plot() method18    # returns a BGR image, so we convert it.19    annotated_image_rgb = annotated_image[..., ::-1]20 21    return annotated_image_rgb22 23 24 25# --- Gradio Interface ---26title = "YOLOv12 Comic Panel Detection"27description = """28This demo showcases a **YOLOv12 object detection model** that has been fine-tuned to detect panels in comic book pages.29Upload an image of a comic page, and the model will draw bounding boxes around each detected panel. 30This can be a useful first step for downstream tasks like Optical Character Recognition (OCR) or character analysis within comics.31"""32 33article = f"""34<div style='text-align: center;'>35    <p style='text-align: center'>Model loaded from <a href='https://huggingface.co/{model_id}' target='_blank'>{model_id}</a></p>36    <p style='text-align: center'>For more details on the training process, check out the project repository: <a href='https://github.com/mosesab/YOLOV12-Comic-Panel-Detection/blob/main/comic-boundary-detection.ipynb' target='_blank'>Comic Boundary Detection</a></p>37    <p>If you like this demo, consider leaving a star on the <a href='https://github.com/mosesab/YOLOV12-Comic-Panel-Detection' target='_blank'>Github repo</a> or a like on the <a href='https://huggingface.co/{model_id}' target='_blank'>Hugging Face model</a>. It helps me know people are interested and motivates further development.</p>38</div>39"""40 41# Define the input components for the Gradio interface42inputs = [43    gr.Image(type="pil", label="Upload Comic Page Image"),44    gr.Slider(45        minimum=0.0,46        maximum=1.0,47        value=0.25, # The default confidence threshold in ultralytics48        step=0.05,49        label="Confidence Threshold",50        info="Filters detections. Only boxes with confidence above this value will be shown."51    ),52    gr.Slider(53        minimum=0.0,54        maximum=1.0,55        value=0.7, # The default IoU threshold in ultralytics56        step=0.05,57        label="IoU Threshold",58        info="Controls merging of overlapping boxes. Higher values allow more overlap."59    )60]61 62examples = [63    ["aura_farmer_1.jpg", 0.25, 0.7],64    ["aura_farmer_2.jpg", 0.25, 0.7],65    ["aura_farmer_3.jpg", 0.25, 0.7],66    ["aura_farmer_4.jpg", 0.25, 0.7],67]68 69gr.Interface(70    fn=detect_panels,71    inputs=inputs,72    outputs=gr.Image(type="pil", label="Detected Panels"),73    title=title,74    description=description,75    article=article,76    examples=examples,77    allow_flagging="auto"78).launch()