ClassCat/DETR-Object-Detection

1 2import torch3from transformers import pipeline4 5from PIL import Image6 7import matplotlib.pyplot as plt8import matplotlib.patches as patches9 10from random import choice11import io12 13detector50 = pipeline(model="facebook/detr-resnet-50")14 15detector101 = pipeline(model="facebook/detr-resnet-101")16 17 18import gradio as gr19 20COLORS = ["#ff7f7f", "#ff7fbf", "#ff7fff", "#bf7fff",21            "#7f7fff", "#7fbfff", "#7fffff", "#7fffbf",22            "#7fff7f", "#bfff7f", "#ffff7f", "#ffbf7f"]23 24fdic = {25    "family" : "Impact",26    "style" : "italic",27    "size" : 15,28    "color" : "yellow",29    "weight" : "bold"30}31 32 33def get_figure(in_pil_img, in_results):34    plt.figure(figsize=(16, 10))35    plt.imshow(in_pil_img)36    #pyplot.gcf()37    ax = plt.gca()38 39    for prediction in in_results:40        selected_color = choice(COLORS)41 42        x, y = prediction['box']['xmin'], prediction['box']['ymin'],43        w, h = prediction['box']['xmax'] - prediction['box']['xmin'], prediction['box']['ymax'] - prediction['box']['ymin']44 45        ax.add_patch(plt.Rectangle((x, y), w, h, fill=False, color=selected_color, linewidth=3))46        ax.text(x, y, f"{prediction['label']}: {round(prediction['score']*100, 1)}%", fontdict=fdic)47 48    plt.axis("off")49 50    return plt.gcf()51 52 53def infer(model, in_pil_img):54 55    results = None56    if model == "detr-resnet-101":57        results = detector101(in_pil_img)58    else:59        results = detector50(in_pil_img)60 61    figure = get_figure(in_pil_img, results)62 63    buf = io.BytesIO()64    figure.savefig(buf, bbox_inches='tight')65    buf.seek(0)66    output_pil_img = Image.open(buf)67 68    return output_pil_img69 70 71with gr.Blocks(title="DETR Object Detection - ClassCat",72                    css=".gradio-container {background:lightyellow;}"73               ) as demo:74    #sample_index = gr.State([])75 76    gr.HTML("""<div style="font-family:'Times New Roman', 'Serif'; font-size:16pt; font-weight:bold; text-align:center; color:royalblue;">DETR Object Detection</div>""")77 78    gr.HTML("""<h4 style="color:navy;">1. Select a model.</h4>""")79 80    model = gr.Radio(["detr-resnet-50", "detr-resnet-101"], value="detr-resnet-50", label="Model name")81 82    gr.HTML("""<br/>""")83    gr.HTML("""<h4 style="color:navy;">2-a. Select an example by clicking a thumbnail below.</h4>""")84    gr.HTML("""<h4 style="color:navy;">2-b. Or upload an image by clicking on the canvas.</h4>""")85 86    with gr.Row():87        input_image = gr.Image(label="Input image", type="pil")88        output_image = gr.Image(label="Output image with predicted instances", type="pil")89 90    gr.Examples(['samples/cats.jpg', 'samples/detectron2.png', 'samples/cat.jpg', 'samples/hotdog.jpg'], inputs=input_image)91 92    gr.HTML("""<br/>""")93    gr.HTML("""<h4 style="color:navy;">3. Then, click "Infer" button to predict object instances. It will take about 10 seconds (on cpu)</h4>""")94 95    send_btn = gr.Button("Infer")96    send_btn.click(fn=infer, inputs=[model, input_image], outputs=[output_image])97 98    gr.HTML("""<br/>""")99    gr.HTML("""<h4 style="color:navy;">Reference</h4>""")100    gr.HTML("""<ul>""")101    gr.HTML("""<li><a href="https://colab.research.google.com/github/facebookresearch/detr/blob/colab/notebooks/detr_attention.ipynb" target="_blank">Hands-on tutorial for DETR</a>""")102    gr.HTML("""</ul>""")103 104 105#demo.queue()106demo.launch(debug=True)107 108 109### EOF ###110