app.py

import torch
import requests
from io import BytesIO
from PIL import Image
import matplotlib.pyplot as plt
import matplotlib.patches as patches
import gradio as gr

from transformers import AutoProcessor, OmDetTurboForObjectDetection

device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"Iniciando no dispositivo: {device.upper()}")

processor = AutoProcessor.from_pretrained("omlab/omdet-turbo-swin-tiny-hf")
model = OmDetTurboForObjectDetection.from_pretrained(
    "omlab/omdet-turbo-swin-tiny-hf"
).to(device)

def plot_results(image, results):
    fig, ax = plt.subplots(1, figsize=(8, 6))
    ax.imshow(image)
    ax.axis("off")

    labels = results.get("text_labels", results.get("classes", []))

    for score, class_name, box in zip(results["scores"], labels, results["boxes"]):
        xmin, ymin, xmax, ymax = box.tolist()
        rect = patches.Rectangle(
            (xmin, ymin),
            xmax - xmin,
            ymax - ymin,
            linewidth=2,
            edgecolor='red',
            facecolor='none'
        )
        ax.add_patch(rect)

        label = f"{class_name}: {score:.2f}"
        ax.text(
            xmin,
            ymin - 5,
            label,
            color='white',
            fontsize=10,
            weight='bold',
            backgroundcolor="red"
        )

    return fig

def detectar_objetos(url, classes_texto):
    try:
        image = Image.open(BytesIO(requests.get(url).content)).convert("RGB")

        classes = [c.strip() for c in classes_texto.split(",")]
        task = "Detect {}.".format(", ".join(classes))

        inputs = processor(
            images=[image],
            text=[classes],
            task=[task],
            return_tensors="pt",
        ).to(device)

        with torch.no_grad():
            outputs = model(**inputs)

        results = processor.post_process_grounded_object_detection(
            outputs,
            text_labels=[classes],
            target_sizes=[image.size[::-1]],
            threshold=0.2,
            nms_threshold=0.3,
        )[0]

        saida = ""
        labels = results.get("text_labels", results.get("classes", []))

        for score, class_name, box in zip(results["scores"], labels, results["boxes"]):
            box_rounded = [round(b, 1) for b in box.tolist()]
            saida += f"{class_name} ({round(score.item(),2)}) -> {box_rounded}\n"

        fig = plot_results(image, results)

        return fig, saida

    except Exception as e:
        return None, f"Erro: {str(e)}"

interface = gr.Interface(
    fn=detectar_objetos,
    inputs=[
        gr.Textbox(label="URL da imagem"),
        gr.Textbox(label="Classes (separadas por vírgula)", value="cat, dog")
    ],
    outputs=[
        gr.Plot(label="Imagem com detecção"),
        gr.Textbox(label="Resultados")
    ],
    title="Detecção de Objetos por URL",
    description="Cole uma URL de imagem e informe os objetos que deseja detectar."
)

interface.launch()