File size: 909 Bytes
72e30d6
 
 
 
 
 
 
 
 
dbb6faf
 
72e30d6
 
 
 
 
 
 
 
 
 
 
 
 
 
5dd94e4
72e30d6
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
from typing import Dict, List, Any
from PIL import Image
from io import BytesIO
from transformers import pipeline
import base64


class EndpointHandler():
    def __init__(self, path=""):
       self.pipeline=pipeline("zero-shot-object-detection",model=path)
        
    def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
        """
       data args:
            images (:obj:`string`)
            candidates (:obj:`list`)
      Return:
            A :obj:`list`:. The list contains items that are dicts should be liked {"label": "XXX", "score": 0.82}
        """
        inputs = data.pop("inputs", data)

        # decode base64 image to PIL
        image = Image.open(BytesIO(base64.b64decode(inputs['image'])))

        # run prediction one image wit provided candiates
        prediction = self.pipeline(image=[image], candidate_labels=inputs["candidates"])
        return prediction[0]