from typing import Dict, List, Any
from transformers import Pipeline
from PIL import Image
from io import BytesIO
import base64
import json
from visual_chatgpt import ImageEditing, Text2Box, Segmenting, Inpainting
import os

class EndpointHandler():
    def __init__(self, path=""):
        # 换个顺序就行了？？？？？离谱
        os.system("ln -s /repository /app/repository")
        self.sam = Segmenting('cuda') 
        self.inpaint = Inpainting('cuda')
        self.grounding = Text2Box('cuda')
        self.model = ImageEditing(self.grounding,self.sam,self.inpaint)
    def __call__(self, data):
        # data=json.loads(data)
        # inputs=data.pop("inputs",data)
        # inputs=base64.b64decode(inputs)
        # raw_images = Image.open(BytesIO(inputs))
        info=data['inputs']
        image=info.pop('image',data)
        image=base64.b64decode(image)
        raw_image=Image.open(BytesIO(image)).convert('RGB')
        texts=info.pop('texts',data)
        target=texts[0]
        replacement=texts[1]
        if replacement=="":
            img=self.model.inference_remove(raw_image,target)
            img_data=img.tobytes()
            img_base64=base64.b64encode(img_data).decode('utf-8')
            return {'image':img_base64}
        else:
            img=self.model.inference_replace_sam(raw_image,target,replacement)
            img_data=img.tobytes()
            img_base64=base64.b64encode(img_data).decode('utf-8')
            return {'image':img_base64}
            
    
if __name__=="__main__":
    my_handler=EndpointHandler(path='.')
    # test_payload={"inputs": "/home/ubuntu/guoling/1.png"}
    with open("/home/ubuntu/guoling/1.png",'rb') as img:
        image_bytes=img.read()
    image_base64=base64.b64encode(image_bytes).decode('utf-8')
    target="the pig"
    replacement=""
    data={
        'inputs':{
            "image":image_base64,
            "target":target,
            "replacement":replacement
        }
    }
    result=my_handler(data)
    result.save("new1.png")