from flask import Flask, request, jsonify from flask_cors import CORS from transformers import pipeline from transformers import AutoProcessor, AutoModelForCausalLM, AutoTokenizer import requests from PIL import Image model_folder_path = r'C:\Users\ravit\OneDrive\Documents\nsfw_filter\nsfw_extension\models' zsc_model_path = r'C:\Users\ravit\OneDrive\Documents\nsfw_filter\nsfw_extension\zscmodel' candidate_labels = ["Violent", "Neutral", "Sexually explicit" ] image_url = r"C:\Users\ravit\OneDrive\Documents\nsfw_filter\nsfw_extension\download.jpeg" print(image_url) tokenizer = AutoTokenizer.from_pretrained(model_folder_path) model = AutoModelForCausalLM.from_pretrained(model_folder_path) processor = AutoProcessor.from_pretrained(model_folder_path) # image = Image.open(requests.get(image_url, stream=True).raw) image = Image.open(image_url) pixel_values = processor(images=image, return_tensors="pt").pixel_values generated_ids = model.generate(pixel_values=pixel_values, max_length=50) generated_caption = processor.batch_decode(generated_ids, skip_special_tokens=True)[0] classifier = pipeline("zero-shot-classification", model=zsc_model_path) output = classifier(generated_caption, candidate_labels, multi_label=False) print(output)