| from transformers import Blip2Processor, Blip2Model, Blip2ForConditionalGeneration | |
| from typing import Dict, List, Any | |
| from PIL import Image | |
| from transformers import pipeline | |
| import requests | |
| import torch | |
| class EndpointHandler(): | |
| def __init__(self, path=""): | |
| """ | |
| path: | |
| """ | |
| self.device = "cuda" if torch.cuda.is_available() else "cpu" | |
| self.processor = Blip2Processor.from_pretrained(path) | |
| self.generate_model = Blip2ForConditionalGeneration.from_pretrained(path, torch_dtype=torch.float16) | |
| self.generate_model.to(self.device) | |
| self.feature_model = Blip2Model.from_pretrained(path, torch_dtype=torch.float16) | |
| self.feature_model.to(self.device) | |
| def __call__(self, data: Dict[str, Any]) -> List[Dict[str, Any]]: | |
| """ | |
| data args: | |
| inputs (:obj: `str` | `PIL.Image` | `np.array`) | |
| kwargs | |
| Return: | |
| A :obj:`list` | `dict`: will be serialized and returned | |
| """ | |
| result = {} | |
| inputs = data.pop("inputs", data) | |
| image_url = inputs['image_url'] | |
| if "prompt" in inputs: | |
| prompt = inputs["prompt"] | |
| else: | |
| prompt = None | |
| if "extract_feature" in inputs: | |
| extract_feature = inputs["extract_feature"] | |
| else: | |
| extract_feature = False | |
| image = Image.open(requests.get(image_url, stream=True).raw) | |
| processed_image = self.processor(images=image, return_tensors="pt").to(self.device, torch.float16) | |
| generated_ids = self.generate_model.generate(**processed_image) | |
| generated_text = self.processor.batch_decode(generated_ids, skip_special_tokens=True)[0].strip() | |
| result["image_caption"] = generated_text | |
| if extract_feature: | |
| caption_feature = self.feature_model(**processed_image) | |
| result["caption_feature"] = caption_feature | |
| if prompt: | |
| prompt_image_processed = self.processor(images=image, text=prompt, return_tensors="pt").to(self.device, torch.float16) | |
| generated_ids = self.generate_model.generate(**prompt_image_processed) | |
| generated_text = self.processor.batch_decode(generated_ids, skip_special_tokens=True)[0].strip() | |
| result["image_prompt"] = generated_text | |
| pass | |
| if extract_feature: | |
| prompt_feature = self.feature_model(**prompt_image_processed) | |
| result["prompt_feature"] = prompt_feature | |
| return result | |