Safetensors
File size: 7,017 Bytes
a197764
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
import re
import torch
import torchvision.transforms as T
from PIL import Image
from torchvision.transforms.functional import InterpolationMode
from transformers import AutoModel, AutoTokenizer


IMAGENET_MEAN = (0.485, 0.456, 0.406)
IMAGENET_STD = (0.229, 0.224, 0.225)

def build_transform(is_train, input_size, pad2square=False, normalize_type='imagenet'):
    if normalize_type == 'imagenet':
        MEAN, STD = IMAGENET_MEAN, IMAGENET_STD
    else:
        raise NotImplementedError
    if is_train:  # use data augumentation
        transform = T.Compose([
            T.Lambda(lambda img: img.convert('RGB') if img.mode != 'RGB' else img),
            T.RandomResizedCrop(input_size, scale=(0.8, 1.0), ratio=(3. / 4., 4. / 3.),
                                interpolation=InterpolationMode.BICUBIC),
            T.ToTensor(),
            T.Normalize(mean=MEAN, std=STD)
        ])
    else:
        transform = T.Compose([
            T.Lambda(lambda img: img.convert('RGB') if img.mode != 'RGB' else img),
            T.Resize((input_size, input_size), interpolation=InterpolationMode.BICUBIC),
            T.ToTensor(),
            T.Normalize(mean=MEAN, std=STD)
        ])

    return transform

def find_closest_aspect_ratio(aspect_ratio, target_ratios, width, height, image_size):
    best_ratio_diff = float('inf')
    best_ratio = (1, 1)
    area = width * height
    for ratio in target_ratios:
        target_aspect_ratio = ratio[0] / ratio[1]
        ratio_diff = abs(aspect_ratio - target_aspect_ratio)
        if ratio_diff < best_ratio_diff:
            best_ratio_diff = ratio_diff
            best_ratio = ratio
        elif ratio_diff == best_ratio_diff:
            if area > 0.5 * image_size * image_size * ratio[0] * ratio[1]:
                best_ratio = ratio
    # print(f'width: {width}, height: {height}, best_ratio: {best_ratio}')
    return best_ratio

def dynamic_preprocess(image, min_num=1, max_num=6, image_size=448, use_thumbnail=False):
    orig_width, orig_height = image.size
    aspect_ratio = orig_width / orig_height

    # calculate the existing image aspect ratio
    target_ratios = set(
        (i, j) for n in range(min_num, max_num + 1) for i in range(1, n + 1) for j in range(1, n + 1) if
        i * j <= max_num and i * j >= min_num)
    target_ratios = sorted(target_ratios, key=lambda x: x[0] * x[1])

    # find the closest aspect ratio to the target
    target_aspect_ratio = find_closest_aspect_ratio(
        aspect_ratio, target_ratios, orig_width, orig_height, image_size)

    # calculate the target width and height
    target_width = image_size * target_aspect_ratio[0]
    target_height = image_size * target_aspect_ratio[1]
    blocks = target_aspect_ratio[0] * target_aspect_ratio[1]

    # resize the image
    resized_img = image.resize((target_width, target_height))
    processed_images = []
    for i in range(blocks):
        box = (
            (i % (target_width // image_size)) * image_size,
            (i // (target_width // image_size)) * image_size,
            ((i % (target_width // image_size)) + 1) * image_size,
            ((i // (target_width // image_size)) + 1) * image_size
        )
        # split the image
        split_img = resized_img.crop(box)
        processed_images.append(split_img)
    assert len(processed_images) == blocks
    if use_thumbnail and len(processed_images) != 1:
        thumbnail_img = image.resize((image_size, image_size))
        processed_images.append(thumbnail_img)
    return processed_images, target_aspect_ratio

def image_process(image_path, config):
    image = Image.open(image_path).convert('RGB')
    transform = build_transform(is_train=False, input_size=config.vision_config.image_size,
                                pad2square=config.pad2square, normalize_type='imagenet')
    if config.dynamic_image_size:
        images, target_aspect_ratio = dynamic_preprocess(image, min_num=config.min_dynamic_patch, max_num=config.max_dynamic_patch,
                                    image_size=config.vision_config.image_size, use_thumbnail=config.use_thumbnail)
    else:
        images = [image]
    pixel_values = [transform(image) for image in images]
    pixel_values = torch.stack(pixel_values)

    return pixel_values.to(torch.bfloat16).cuda(), torch.tensor([[target_aspect_ratio[0], target_aspect_ratio[1]]], dtype=torch.long)

def parse_block_pos(str_, target_aspect_ratio):
    block_num_w, block_num_h = target_aspect_ratio[0][0], target_aspect_ratio[0][1]
    action, location, direction, location_or_text = None, None, None, None
    str_ = str_.strip()
    match = re.match(r'^(.*?)\((.*?)\)$', str_)

    if match:
        action, location_or_text = match.groups()

        if action == 'CLICK':
            match = re.match(r'^\[(\d{1}), (\d{3}), (\d{3})\].*?$', location_or_text)

            if match:
                block_idx, cx, cy = match.groups()

                block_idx = int(block_idx)
                cx = int(cx)
                cy = int(cy)

                cx += (block_idx % block_num_w) * 1000
                cy += (block_idx // block_num_w) * 1000
                cx /= block_num_w * 1000
                cy /= block_num_h * 1000

                location = [cx, cy]
            else:
                print(location_or_text)

        elif action.startswith('SWIPE_'):
            action, direction = action.split('_', 1)

    return {
        'action': action,
        'location': location,
        'direction': direction,
        'content': location_or_text
    }

question_template = '''## Task: {task}
## History Actions:
{history}
## Action Space
1. CLICK([block_index, cx, cy], "text")
2. TYPE("text")
3. PRESS_BACK()
4. PRESS_HOME()
5. PRESS_ENTER()
6. SWIPE_UP()
7. SWIPE_DOWN()
8. SWIPE_LEFT()
9. SWIPE_RIGHT()
10. COMPLETED()
## Requirements: Please infer the next action according to the Task and History Actions. Think step by step. Return with Image Description, Next Action Description and Action Code. The Action Code should follow the definition in the Action Space.'''


path = './SpiritSight-Agent-2B-base'
model = AutoModel.from_pretrained(
    path,
    torch_dtype=torch.bfloat16,
    low_cpu_mem_usage=True,
    # use_flash_attn=False,
    trust_remote_code=True).eval().cuda()
tokenizer = AutoTokenizer.from_pretrained(path, trust_remote_code=True, use_fast=False)

task = "Go to search bar in Google Chrome then search for walmart."
history = ""
question = question_template.format(task=task, history=history)

image_path = './image.png'
pixel_values, target_aspect_ratio = image_process(image_path, model.config)
generation_config = dict(max_new_tokens=1024, do_sample=True)
response = model.chat(
    tokenizer=tokenizer,
    pixel_values=pixel_values,
    question=question,
    target_aspect_ratio=target_aspect_ratio,
    generation_config=generation_config
)
print(response)

action_code_str = response.split()[-1]
action_code = parse_block_pos(action_code_str, target_aspect_ratio.cpu().numpy())
print(action_code)