File size: 2,279 Bytes
24c4def
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# import sys
# sys.path.append("/home/wcx/wcx/EasyDetect/pipeline")
from pipeline.claim_generate import * 
from pipeline.query_generate import *
from pipeline.tool_execute import *
from pipeline.judge import *
from pipeline.openai_wrapper import *

class Pipeline:
    def __init__(self):
        # 全局只实例化一个对象 会不会干扰prompt的结果
        self.syncchat = SyncChat(model="gpt-4-1106-preview", api_key="sk-jD8DeGdJKrdOxpiQ5bD4845bB53346C3A0E9Ed479bE08676", base_url="https://oneapi.xty.app/v1")
        self.asyncchat = AsyncChat(model="gpt-4-1106-preview", api_key="sk-jD8DeGdJKrdOxpiQ5bD4845bB53346C3A0E9Ed479bE08676", base_url="https://oneapi.xty.app/v1")
        self.visionchat = VisionChat(model="gpt-4-vision-preview", api_key="sk-jD8DeGdJKrdOxpiQ5bD4845bB53346C3A0E9Ed479bE08676", base_url="https://oneapi.xty.app/v1")

        self.claim_generator = ClaimGenerator(prompt_path="/home/wcx/wcx/EasyDetect/prompts/claim_generate.yaml",chat=self.syncchat)
        self.query_generator = QueryGenerator(prompt_path="/home/wcx/wcx/EasyDetect/prompts/query_generate.yaml",chat=self.asyncchat, type="image-to-text")
        self.tool = Tool()
        self.judger = Judger(prompt_path="/home/wcx/wcx/EasyDetect/prompts/verification.yaml", chat=self.visionchat, type="image-to-text")

    def run(self, text, image_path):
        response, claim_list = self.claim_generator.get_response(text=text)
        objects, attribute_ques_list, scenetext_ques_list, fact_ques_list = self.query_generator.get_response(claim_list=claim_list)
        object_res, attribue_res, text_res, fact_res = self.tool.execute(image_path=image_path,
                                                            new_path="/newdisk3/wcx/MLLM/image-to-text/cache", 
                                                            objects=objects,
                                                            attribute_list=attribute_ques_list, 
                                                            scenetext_list=scenetext_ques_list, 
                                                            fact_list=fact_ques_list)

        # response = self.judger.get_response(object_res, attribue_res, text_res, fact_res, claim_list, image_path)
        return object_res["phrases"]