app.py

from transformers import M2M100ForConditionalGeneration, M2M100Tokenizer
from turtle import title
from transformers import pipeline
import numpy as np
from PIL import Image
import sys, os
import gradio as gr




model = M2M100ForConditionalGeneration.from_pretrained("facebook/m2m100_418M")
tokenizer = M2M100Tokenizer.from_pretrained("facebook/m2m100_418M")

pipe = pipeline("zero-shot-image-classification", model="openai/clip-vit-base-patch32")
images="dog.jpg"

def shot(image, labels_text):

    
    
    PIL_image = Image.fromarray(np.uint8(image)).convert('RGB')
    labels = labels_text.split(",")
    
   
    # Translate 
    
    res = pipe(images=PIL_image, 
           candidate_labels=labels,
           hypothesis_template= "This is a photo of a {}")
    
    tokenizer.src_lang = "en"
    encodedText = tokenizer(labels_text, return_tensors="pt")
    generatedTokens = model.generate(**encodedText, forced_bos_token_id=tokenizer.get_lang_id("ru"))    
    return tokenizer.batch_decode(generatedTokens, skip_special_tokens=True)[0], {dic["label"]: dic["score"] for dic in res}
     
    
iface = gr.Interface(shot, 
                    ["image", "text"], 
                    "label", 
                    examples=[["dog.jpg", "dog,cat,bird"]],                         
                    description="Add a picture and a list of labels separated by commas",
                    title="Zero-shot Image Classification")

iface.launch()