app.py

import gradio as gr
import supervision as sv
from PIL import Image
from rfdetr import RFDETRMedium, RFDETRSegPreview
from rfdetr.detr import RFDETR
from rfdetr.util.coco_classes import COCO_CLASSES

MARKDOWN = """
# (RF-DETR Demo-Alliance Bioversity-CIAT)
RF-DETR object detection and segmentation
"""

COLOR = sv.ColorPalette.from_hex([
    "#ffff00", "#ff9b00", "#ff8080", "#ff66b2", "#ff66ff", "#b266ff",
    "#9999ff", "#3399ff", "#66ffff", "#33ff99", "#66ff66", "#99ff00"
])

def load_model(resolution: int, checkpoint: str) -> RFDETR:
    if checkpoint == "medium (object detection)":
        return RFDETRMedium(resolution=resolution)
    if checkpoint == "segmentation preview":
        return RFDETRSegPreview(resolution=resolution)
    raise TypeError("Checkpoint must be medium (object detection) or segmentation preview.")

def adjust_resolution(checkpoint: str, resolution: int) -> int:
    if checkpoint == "segmentation preview":
        divisor = 24
    elif checkpoint == "medium (object detection)":
        divisor = 32
    else:
        raise ValueError(f"Unknown checkpoint: {checkpoint}")
    remainder = resolution % divisor
    if remainder == 0:
        return resolution
    lower = resolution - remainder
    upper = lower + divisor
    if resolution - lower < upper - resolution:
        return lower
    else:
        return upper

def image_processing_inference(
    input_image: Image.Image,
    confidence: float,
    resolution: int,
    checkpoint: str
):
    if input_image is None:
        return None
    resolution = adjust_resolution(checkpoint=checkpoint, resolution=resolution)
    model = load_model(resolution=resolution, checkpoint=checkpoint)
    detections = model.predict(input_image, threshold=confidence)

    resolution_wh = (input_image.width, input_image.height)
    text_scale = sv.calculate_optimal_text_scale(resolution_wh=resolution_wh) - 0.2
    thickness = sv.calculate_optimal_line_thickness(resolution_wh=resolution_wh)

    mask_annotator = sv.MaskAnnotator(color=COLOR)
    bbox_annotator = sv.BoxAnnotator(color=COLOR, thickness=thickness)
    label_annotator = sv.LabelAnnotator(
        color=COLOR,
        text_color=sv.Color.BLACK,
        text_scale=text_scale
    )

    labels = [
        f"{COCO_CLASSES[class_id]} {confidence:.2f}"
        for class_id, confidence
        in zip(detections.class_id, detections.confidence)
    ]

    annotated_image = input_image.copy()
    annotated_image = bbox_annotator.annotate(annotated_image, detections)
    annotated_image = label_annotator.annotate(annotated_image, detections, labels)
    if checkpoint == "segmentation preview":
        annotated_image = mask_annotator.annotate(annotated_image, detections)
    return annotated_image

with gr.Blocks() as demo:
    gr.Markdown(MARKDOWN)
    with gr.Row():
        image_processing_input_image = gr.Image(
            label="Upload image",
            image_mode="RGB",
            type="pil",
            height=600
        )
        image_processing_output_image = gr.Image(
            label="Output image",
            image_mode="RGB",
            type="pil",
            height=600
        )
    with gr.Row():
        with gr.Column():
            image_processing_confidence_slider = gr.Slider(
                label="Confidence",
                minimum=0.0,
                maximum=1.0,
                step=0.05,
                value=0.5,
            )
            image_processing_resolution_slider = gr.Slider(
                label="Inference resolution",
                minimum=224,
                maximum=2240,
                step=1,
                value=896,
            )
            image_processing_checkpoint_dropdown = gr.Dropdown(
                label="Checkpoint",
                choices=[
                    "medium (object detection)",
                    "segmentation preview"
                ],
                value="medium (object detection)"
            )
        with gr.Column():
            image_processing_submit_button = gr.Button("Submit")

    image_processing_submit_button.click(
        image_processing_inference,
        inputs=[
            image_processing_input_image,
            image_processing_confidence_slider,
            image_processing_resolution_slider,
            image_processing_checkpoint_dropdown
        ],
        outputs=image_processing_output_image,
    )

if __name__ == "__main__":
    demo.launch(debug=False, show_error=True)