File size: 14,261 Bytes
032e687
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
import json
import os.path

from mmengine.model import BaseModel

from lmdeploy import pipeline, TurbomindEngineConfig
from lmdeploy.vl import load_image
import torch
import torch.nn as nn

class LLM_Annotor(BaseModel):
    def __init__(self,
        model=None,
        save_folder='./work_dirs/qwen2_72b_obj_referring/'
        ):
        super().__init__()
        print(torch.cuda.device_count())
        print(f"\n\n Using {model} !!! \n\n")
        pipe = pipeline(model,
                        backend_config=TurbomindEngineConfig(
                            # session_len=8192,
                            session_len=4096,
                            tp=torch.cuda.device_count()),
                        )
        self.pipe = [pipe]
        self._zero = nn.Linear(10, 10)
        self.results_list = []
        self.item_idx = 0

        if not os.path.exists(save_folder):
            os.mkdir(save_folder)
        self.save_folder = save_folder

    def forward(self, **kwargs):
        return None

    def predict_forward_text_vertify(self, data_dicts):
        prompts = []

        print('vertify forward !!!')

        # remove ignore items
        if 'ignore' in data_dicts[0].keys():
            data_dicts_ = []
            for _item in data_dicts:
                if _item['ignore']:
                    continue
                data_dicts_.append(_item)
            data_dicts = data_dicts_

        for data_dict in data_dicts:
            texts = data_dict['text_prompt']
            prompts.append(texts)

        response_list = self.pipe[0](prompts)
        text_lsit = [item.text for item in response_list]

        results_list = []

        for i, text in enumerate(text_lsit):
            print('\n\n', text, '\n\n')
            if 'No conflict' in text:
                results_list.append({
                    'video_id': data_dicts[i]['video_id'],
                    'obj_id': data_dicts[i]['obj_id'],
                    'ori_captions': data_dicts[i]['ori_captions'],
                })
                print('\n\n', data_dicts[i]['ori_captions'], '\n\n')

        self.results_list += results_list

        if len(self.results_list) > 100:
            self.save_step()

        return {}

    def predict_forward_text_summarize(self, data_dicts):
        prompts = []

        print('summarize forward !!!')

        for data_dict in data_dicts:
            texts = data_dict['text_prompt']
            prompts.append(texts)

        response_list = self.pipe[0](prompts)
        text_lsit = [item.text for item in response_list]

        results_list = []

        for i, text in enumerate(text_lsit):

            results_list.append({
                'video_id': data_dicts[i]['video_id'],
                'obj_id': data_dicts[i]['obj_id'],
                'ori_captions': data_dicts[i]['ori_captions'],
                'summarized': text
            })
            print('\n\n', data_dicts[i]['ori_captions'], '\n', text, '\n\n')

        self.results_list += results_list

        if len(self.results_list) > 100:
            self.save_step()

        return {}

    def predict_forward_text_formatting(self, data_dicts):
        prompts = []

        print('formatting forward !!!')

        for data_dict in data_dicts:
            texts = data_dict['text_prompt']
            prompts.append(texts)

        response_list = self.pipe[0](prompts)
        text_lsit = [item.text for item in response_list]

        results_list = []

        for i, text in enumerate(text_lsit):
            text = text.split(':')[-1]
            text = text.strip()
            text = text.replace('\"', '')
            results_list.append({
                'video_id': data_dicts[i]['video_id'],
                'obj_id': data_dicts[i]['obj_id'],
                'ori_captions': data_dicts[i]['ori_captions'],
                'summarized': data_dicts[i]['summarized'],
                'formated': text
            })
            print('\n\n', text, '\n\n')

        self.results_list += results_list

        if len(self.results_list) > 100:
            self.save_step()

        return {}

    def predict_forward_short_cap(self, data_dicts):
        prompts = []

        print('short caption forward !!!')

        for data_dict in data_dicts:
            texts = data_dict['text_prompt']
            prompts.append(texts)

        response_list = self.pipe[0](prompts)
        text_lsit = [item.text for item in response_list]

        results_list = []

        for i, text in enumerate(text_lsit):
            print('\n\n', data_dicts[i]['formated'], '\n', text, '\n\n')
            results_list.append({
                'video_id': data_dicts[i]['video_id'],
                'obj_id': data_dicts[i]['obj_id'],
                'formated': data_dicts[i]['formated'],
                'short_cap': text
            })

        self.results_list += results_list

        if len(self.results_list) > 100:
            self.save_step()

        return {}

    def predict_forward_filter_unindentified(self, data_dicts):
        prompts = []

        print('filter_unindentified forward !!!')

        for data_dict in data_dicts:
            texts = data_dict['text_prompt']
            prompts.append(texts)

        response_list = self.pipe[0](prompts)
        text_lsit = [item.text for item in response_list]

        results_list = []

        for i, text in enumerate(text_lsit):
            # print('\n\n', data_dicts[i]['caption'], '\n', text, '\n\n')
            if "Unidentified" in text or "unidentified" in text:
                pass
            else:
                results_list.append({
                    'video_id': data_dicts[i]['video_id'],
                    'obj_id': data_dicts[i]['obj_id'],
                    'caption': data_dicts[i]['caption'],
                    'type': data_dicts[i]['type'],
                    'category': text,
                })

        self.results_list += results_list
        if len(self.results_list) > 100:
            self.save_step()
        return {}

    def predict_forward_consistency(self, data_dicts):
        prompts = []

        print('Consistency forward !!!')

        for data_dict in data_dicts:
            texts = data_dict['text_prompt']
            prompts.append(texts)

        response_list = self.pipe[0](prompts)
        text_lsit = [item.text for item in response_list]

        results_list = []

        out_num = 0
        for i, text in enumerate(text_lsit):
            if "Yes" in text or "yes" in text:
                print('\n\n', data_dicts[i]['text_prompt'], '\n', text, '\n\n')
                out_num += 1
                results_list.append({
                    'video_id': data_dicts[i]['video_id'],
                    'obj_id': data_dicts[i]['obj_id'],
                    'caption': data_dicts[i]['caption'],
                    'category': data_dicts[i]['category'],
                })

        print(f"***************Input {len(text_lsit)} items and keep {out_num} items !!!\n")
        self.results_list += results_list
        if len(self.results_list) > 100:
            self.save_step()
        return {}

    def predict_forward_re_consistency(self, data_dicts):
        prompts = []

        print('Re consistency forward !!!')

        for data_dict in data_dicts:
            texts = data_dict['text_prompt']
            prompts.append(texts)

        response_list = self.pipe[0](prompts)
        text_lsit = [item.text for item in response_list]

        results_list = []

        out_num = 0
        for i, text in enumerate(text_lsit):
            if "Yes" in text or "yes" in text:
                # print('\n\n', data_dicts[i]['text_prompt'], '\n', text, '\n\n')
                out_num += 1
                results_list.append({
                    'video_id': data_dicts[i]['video_id'],
                    'obj_id': data_dicts[i]['obj_id'],
                    'crop_caption': data_dicts[i]['crop_caption'],
                    'crop_category': data_dicts[i]['crop_category'],
                    'image_caption': data_dicts[i]['image_caption'],
                    'video_caption': data_dicts[i]['video_caption'],
                })
            # else:
            #     print('\n\n', data_dicts[i]['text_prompt'], '\n', text, '\n\n')
        print(f"***************Input {len(text_lsit)} items and keep {out_num} items !!!\n")
        self.results_list += results_list
        if len(self.results_list) > 100:
            self.save_step()
        return {}

    def predict_forward_change_style(self, data_dicts):
        prompts = []

        print('Change Style forward !!!')

        for data_dict in data_dicts:
            texts = data_dict['text_prompt']
            prompts.append(texts)

        response_list = self.pipe[0](prompts)
        text_lsit = [item.text for item in response_list]

        results_list = []

        for i, text in enumerate(text_lsit):
            results_list.append({
                'video_id': data_dicts[i]['video_id'],
                'obj_id': data_dicts[i]['obj_id'],
                'crop_caption': data_dicts[i]['crop_caption'],
                'crop_category': data_dicts[i]['crop_category'],
                'image_caption': data_dicts[i]['image_caption'],
                'video_caption': data_dicts[i]['video_caption'],
                'final_caption': text,
            })
            # print('\n\n', data_dicts[i]['text_prompt'], '\n', text, '\n\n')
        self.results_list += results_list
        if len(self.results_list) > 100:
            self.save_step()
        return {}

    def predict_forward_translation(self, data_dicts):
        prompts = []

        print('translation forward !!!')

        for data_dict in data_dicts:
            texts = data_dict['text_prompt']
            prompts.append(texts)

        response_list = self.pipe[0](prompts)
        text_lsit = [item.text for item in response_list]

        results_list = []

        for i, text in enumerate(text_lsit):
            results_list.append({
                'video_id': data_dicts[i]['video_id'],
                'obj_id': data_dicts[i]['obj_id'],
                'crop_caption': data_dicts[i]['crop_caption'],
                'crop_category': data_dicts[i]['crop_category'],
                'image_caption': data_dicts[i]['image_caption'],
                'video_caption': data_dicts[i]['video_caption'],
                'final_caption': data_dicts[i]['final_caption'],
                'translation': text
            })
            print(text, '\n')
            # print('\n\n', data_dicts[i]['text_prompt'], '\n', text, '\n\n')
        self.results_list += results_list
        if len(self.results_list) > 100:
            self.save_step()
        return {}

    def predict_forward_text(self, data_dicts):
        if 'task' in data_dicts[0].keys() and data_dicts[0]['task'] == 'vertify':
            return self.predict_forward_text_vertify(data_dicts)

        prompts = []

        # remove ignore items
        if 'ignore' in data_dicts[0].keys():
            data_dicts_ = []
            for _item in data_dicts:
                if _item['ignore']:
                    continue
                data_dicts_.append(_item)
            data_dicts = data_dicts_

        for data_dict in data_dicts:
            texts = data_dict['text_prompt']
            prompts.append(texts)

        response_list = self.pipe[0](prompts)
        text_lsit = [item.text for item in response_list]

        results_list = []

        for i, text in enumerate(text_lsit):
            results_list.append({
                'video_id': data_dicts[i]['video_id'],
                'obj_id': data_dicts[i]['obj_id'],
                'caption': text,
            })
            print('\n\n', text, '\n\n')

        self.results_list += results_list

        if len(self.results_list) > 100:
            self.save_step()

        return {}

    def save_step(self, last=False):
        if last:
            save_list = self.results_list
        else:
            save_list = self.results_list[:100]
            self.results_list = self.results_list[100:]

        json_path = os.path.join(self.save_folder, f'{self.item_idx}.json')
        self.item_idx += 1
        with open(json_path, 'w') as f:
            json.dump(save_list, fp=f)
        return

    def predict_forward(self, image_paths, **kwargs):

        if 'type' in kwargs.keys() and kwargs['type'] == 'text':
            if 'task' in kwargs.keys() and kwargs['task'] == 'vertify':
                return self.predict_forward_text_vertify(kwargs['data_dicts'])
            elif 'task' in kwargs.keys() and kwargs['task'] == 'summarize':
                return self.predict_forward_text_summarize(kwargs['data_dicts'])
            elif 'task' in kwargs.keys() and kwargs['task'] == 'formatting':
                return self.predict_forward_text_formatting(kwargs['data_dicts'])
            elif 'task' in kwargs.keys() and kwargs['task'] == 'short_cap':
                return self.predict_forward_short_cap(kwargs['data_dicts'])
            elif 'task' in kwargs.keys() and kwargs['task'] == 'filter_unindentified':
                return self.predict_forward_filter_unindentified(kwargs['data_dicts'])
            elif 'task' in kwargs.keys() and kwargs['task'] == 'consistency':
                return self.predict_forward_consistency(kwargs['data_dicts'])
            elif 'task' in kwargs.keys() and kwargs['task'] == 're_consistency':
                return self.predict_forward_re_consistency(kwargs['data_dicts'])
            elif 'task' in kwargs.keys() and kwargs['task'] == 'change_style':
                return self.predict_forward_change_style(kwargs['data_dicts'])
            elif 'task' in kwargs.keys() and kwargs['task'] == 'translation':
                return self.predict_forward_translation(kwargs['data_dicts'])
            return self.predict_forward_text(kwargs['data_dicts'])

        images = [load_image(image_path) for image_path in image_paths]
        prompts = [('Please briefly describe this image in a sentence.', image) for image in images]
        response_list = self.pipe[0](prompts)
        text_lsit = [item.text for item in response_list]
        print(text_lsit)
        return {}