File size: 4,689 Bytes
7fa80e9
de6a4a2
 
 
 
 
 
 
 
7fa80e9
 
de6a4a2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
560a2a7
de6a4a2
 
560a2a7
de6a4a2
 
 
 
 
560a2a7
de6a4a2
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f3a13dd
de6a4a2
 
 
 
 
 
 
 
9b70520
de6a4a2
 
13a0d78
de6a4a2
 
 
 
 
 
 
 
 
 
 
398c325
de6a4a2
398c325
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
de6a4a2
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
import gradio as gr
import spaces
from transformers import Qwen2VLForConditionalGeneration, AutoTokenizer, AutoProcessor
from qwen_vl_utils import process_vision_info
import torch
import base64
from PIL import Image, ImageDraw
from io import BytesIO
import re


models = {
    "OS-Copilot/OS-Atlas-Base-7B": Qwen2VLForConditionalGeneration.from_pretrained("OS-Copilot/OS-Atlas-Base-7B", torch_dtype="auto", device_map="auto"),
}

processors = {
    "OS-Copilot/OS-Atlas-Base-7B": AutoProcessor.from_pretrained("OS-Copilot/OS-Atlas-Base-7B")
}


def image_to_base64(image):
    buffered = BytesIO()
    image.save(buffered, format="PNG")
    img_str = base64.b64encode(buffered.getvalue()).decode("utf-8")
    return img_str


def draw_bounding_boxes(image, bounding_boxes, outline_color="red", line_width=2):
    draw = ImageDraw.Draw(image)
    for box in bounding_boxes:
        xmin, ymin, xmax, ymax = box
        draw.rectangle([xmin, ymin, xmax, ymax], outline=outline_color, width=line_width)
    return image


def rescale_bounding_boxes(bounding_boxes, original_width, original_height, scaled_width=1000, scaled_height=1000):
    x_scale = original_width / scaled_width
    y_scale = original_height / scaled_height
    rescaled_boxes = []
    for box in bounding_boxes:
        xmin, ymin, xmax, ymax = box
        rescaled_box = [
            xmin * x_scale,
            ymin * y_scale,
            xmax * x_scale,
            ymax * y_scale
        ]
        rescaled_boxes.append(rescaled_box)
    return rescaled_boxes


@spaces.GPU
def run_example(image, text_input, model_id="OS-Copilot/OS-Atlas-Base-7B"):
    model = models[model_id].eval()
    processor = processors[model_id]
    prompt = f"In this UI screenshot, what is the position of the element corresponding to the command \"{text_input}\" (with bbox)?"
    messages = [
        {
            "role": "user",
            "content": [
                {"type": "image", "image": f"data:image;base64,{image_to_base64(image)}"},
                {"type": "text", "text": prompt},
            ],
        }
    ]

    text = processor.apply_chat_template(
        messages, tokenize=False, add_generation_prompt=True
    )
    image_inputs, video_inputs = process_vision_info(messages)
    inputs = processor(
        text=[text],
        images=image_inputs,
        videos=video_inputs,
        padding=True,
        return_tensors="pt",
    )
    inputs = inputs.to("cuda")

    generated_ids = model.generate(**inputs, max_new_tokens=128)
    generated_ids_trimmed = [
        out_ids[len(in_ids) :] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
    ]
    output_text = processor.batch_decode(
        generated_ids_trimmed, skip_special_tokens=False, clean_up_tokenization_spaces=False
    )
    print(output_text)
    text = output_text[0]

    object_ref_pattern = r"<\|object_ref_start\|>(.*?)<\|object_ref_end\|>"
    box_pattern = r"<\|box_start\|>(.*?)<\|box_end\|>"

    object_ref = re.search(object_ref_pattern, text).group(1)
    box_content = re.search(box_pattern, text).group(1)

    boxes = [tuple(map(int, pair.strip("()").split(','))) for pair in box_content.split("),(")]
    boxes = [[boxes[0][0], boxes[0][1], boxes[1][0], boxes[1][1]]]

    scaled_boxes = rescale_bounding_boxes(boxes, image.width, image.height)
    return object_ref, draw_bounding_boxes(image, scaled_boxes)

css = """
  #output {
    height: 500px; 
    overflow: auto; 
    border: 1px solid #ccc; 
  }
"""
with gr.Blocks(css=css) as demo:
    gr.Markdown(
    """
    # Demo for OS-ATLAS: A Foundation Action Model For Generalist GUI Agents
    """)
    with gr.Row():
        with gr.Column():
            input_img = gr.Image(label="Input Image", type="pil")
            model_selector = gr.Dropdown(choices=list(models.keys()), label="Model", value="OS-Copilot/OS-Atlas-Base-7B")
            text_input = gr.Textbox(label="User Prompt")
            submit_btn = gr.Button(value="Submit")
        with gr.Column():
            model_output_text = gr.Textbox(label="Model Output Text")
            annotated_image = gr.Image(label="Annotated Image")

    gr.Examples(
        examples=[
            ["assets/web_6f93090a-81f6-489e-bb35-1a2838b18c01.png", "select search textfield"],
            ["assets/web_6f93090a-81f6-489e-bb35-1a2838b18c01.png", "switch to discussions"],
        ],
        inputs=[input_img, text_input],
        outputs=[model_output_text, annotated_image],
        fn=run_example,
        cache_examples=True,
        label="Try examples"
    )

    submit_btn.click(run_example, [input_img, text_input, model_selector], [model_output_text, annotated_image])

demo.launch(debug=True)