handler.py

from typing import Dict, List, Any
import numpy as np
from transformers import CLIPProcessor, CLIPModel
from PIL import Image
from io import BytesIO
import base64
import requests

# handle clip embeddings by utilizing openAI CLIP pretrained model
class EndpointHandler():
    def __init__(self, path=""):
        # Preload all the elements you we need at inference.
        self.model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32")
        self.processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")


    def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]:
        # inputs = self.processor(text=["a photo of a cat", "a photo of a dog"], images=image, return_tensors="pt", padding=True)
        # logits_per_image = outputs.logits_per_image # this is the image-text similarity score
        # probs = logits_per_image.softmax(dim=1) # we can take the softmax to get the label probabilities
        # print(probs)
        
        # data = {
        #     "inputs": {
        #         "image": "http://images.cocodataset.org/val2017/000000039769.jpg",
        #         "text": ["a photo of a cat", "a photo of a dog"]
        #     },
        # }
        
        inputs = data.get("inputs")
        imageBase64 = inputs.get("image")
        # imageURL = inputs.get("image")
        text = inputs.get("text")
        # print(imageURL)
        # print(text)
        # image = Image.open(requests.get(imageBase64, stream=True).raw)
        
        image = Image.open(BytesIO(base64.b64decode(imageBase64.split(",")[1].encode())))
        inputs = self.processor(text=text, images=image, return_tensors="pt", padding=True)
        outputs = self.model(**inputs)
        embeddings = outputs.image_embeds.detach().numpy().flatten().tolist()
        return { "embeddings": embeddings }