File size: 5,824 Bytes
14a799a
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import ctypes
import os
from typing import List, Tuple
import numpy as np
import platform
from pyaxdev import _lib, AxDeviceType, AxDevices, check_error


class ClipInit(ctypes.Structure):
    _fields_ = [
        ('dev_type', AxDeviceType),
        ('devid', ctypes.c_char),
        ('text_encoder_path', ctypes.c_char * 128),
        ('image_encoder_path', ctypes.c_char * 128),
        ('tokenizer_path', ctypes.c_char * 128),
        ('isCN', ctypes.c_char),
        ('db_path', ctypes.c_char * 128)
    ]

class ClipImage(ctypes.Structure):
    _fields_ = [
        ('data', ctypes.POINTER(ctypes.c_ubyte)),
        ('width', ctypes.c_int),
        ('height', ctypes.c_int),
        ('channels', ctypes.c_int),
        ('stride', ctypes.c_int)
    ]

class ClipFeatureItem(ctypes.Structure):
    _fields_ = [
        ('feat', ctypes.c_float * 768),
        ('len', ctypes.c_int)
    ]


class ClipResultItem(ctypes.Structure):
    _fields_ = [
        ('key', ctypes.c_char * 64),
        ('score', ctypes.c_float)
    ]

_lib.clip_create.argtypes = [ctypes.POINTER(ClipInit), ctypes.POINTER(ctypes.c_void_p)]
_lib.clip_create.restype = ctypes.c_int

_lib.clip_destroy.argtypes = [ctypes.c_void_p]
_lib.clip_destroy.restype = ctypes.c_int

_lib.clip_add.argtypes = [ctypes.c_void_p, ctypes.c_char_p, ctypes.POINTER(ClipImage), ctypes.c_char]
_lib.clip_add.restype = ctypes.c_int

_lib.clip_remove.argtypes = [ctypes.c_void_p, ctypes.c_char_p]
_lib.clip_remove.restype = ctypes.c_int

_lib.clip_contain.argtypes = [ctypes.c_void_p, ctypes.c_char_p]
_lib.clip_contain.restype = ctypes.c_int

_lib.clip_get_text_feat.argtypes = [ctypes.c_void_p, ctypes.c_char_p, ctypes.POINTER(ClipFeatureItem)]
_lib.clip_get_text_feat.restype = ctypes.c_int

_lib.clip_match_feat.argtypes = [ctypes.c_void_p, ctypes.POINTER(ClipFeatureItem), ctypes.POINTER(ClipResultItem), ctypes.c_int]
_lib.clip_match_feat.restype = ctypes.c_int

_lib.clip_match_text.argtypes = [ctypes.c_void_p, ctypes.c_char_p, ctypes.POINTER(ClipResultItem), ctypes.c_int]
_lib.clip_match_text.restype = ctypes.c_int

_lib.clip_match_image.argtypes = [ctypes.c_void_p, ctypes.POINTER(ClipImage), ctypes.POINTER(ClipResultItem), ctypes.c_int]
_lib.clip_match_image.restype = ctypes.c_int


class Clip:
    def __init__(self, init_info: dict):
        self.handle = None
        self.init_info = ClipInit()
        
        # 设置初始化参数
        self.init_info.dev_type = init_info.get('dev_type', AxDeviceType.axcl_device)
        self.init_info.devid = init_info.get('devid', 0)
        self.init_info.isCN = init_info.get('isCN', 1)
        
        # 设置路径
        for path_name in ['text_encoder_path', 'image_encoder_path', 'tokenizer_path', 'db_path']:
            if path_name in init_info:
                setattr(self.init_info, path_name, init_info[path_name].encode('utf-8'))
        
        # 创建CLIP实例
        handle = ctypes.c_void_p()
        check_error(_lib.clip_create(ctypes.byref(self.init_info), ctypes.byref(handle)))
        self.handle = handle

    def __del__(self):
        if self.handle:
            _lib.clip_destroy(self.handle)

    def add_image(self, key: str, image_data: np.ndarray) -> None:
        if self.contains_image(key):
            return
        image = ClipImage()
        image.data = ctypes.cast(image_data.ctypes.data, ctypes.POINTER(ctypes.c_ubyte))
        image.width = image_data.shape[1]
        image.height = image_data.shape[0]
        image.channels = image_data.shape[2]
        image.stride = image_data.shape[1] * image_data.shape[2]
        
        check_error(_lib.clip_add(self.handle, key.encode('utf-8'), ctypes.byref(image), 0))

    def remove_image(self, key: str) -> None:
        check_error(_lib.clip_remove(self.handle, key.encode('utf-8')))

    def contains_image(self, key: str) -> bool:
        return _lib.clip_contain(self.handle, key.encode('utf-8')) == 1

    def get_text_feat(self, text: str) -> np.ndarray:
        feat = ClipFeatureItem()
        check_error(_lib.clip_get_text_feat(self.handle, text.encode('utf-8'), ctypes.byref(feat)))
        return np.array(feat.feat[:feat.len])

    def match_feat(self, feat: np.ndarray, top_k: int = 10) -> List[Tuple[str, float]]:
        feat_item = ClipFeatureItem()
        # feat_item.feat = feat.astype(np.float32).tolist()
        arr = feat.astype(np.float32)
        for i in range(len(arr)):
            feat_item.feat[i] = arr[i]
        feat_item.len = len(feat)
        
        results = (ClipResultItem * top_k)()
        check_error(_lib.clip_match_feat(self.handle, ctypes.byref(feat_item), results, top_k))
        
        return [(item.key.decode('utf-8'), item.score) for item in results]
    
    def match_text(self, text: str, top_k: int = 10) -> List[Tuple[str, float]]:
        results = (ClipResultItem * top_k)()
        check_error(_lib.clip_match_text(self.handle, text.encode('utf-8'), results, top_k))
        
        return [(item.key.decode('utf-8'), item.score) for item in results]

    def match_image(self, image_data: bytes, width: int, height: int, channels: int = 3, top_k: int = 10) -> List[Tuple[str, float]]:
        image = ClipImage()
        image.data = ctypes.cast(ctypes.create_string_buffer(image_data), ctypes.POINTER(ctypes.c_ubyte))
        image.width = width
        image.height = height
        image.channels = channels
        image.stride = width * channels
        
        results = (ClipResultItem * top_k)()
        check_error(_lib.clip_match_image(self.handle, ctypes.byref(image), ctypes.byref(results), top_k))
        
        return [(item.key.decode('utf-8'), item.score) for item in results]