app.py

import os
import time
import gradio as gr
import torch
from PIL import Image
from gtts import gTTS
import numpy as np
import cv2
from transformers import BlipProcessor, BlipForConditionalGeneration, AutoTokenizer, AutoModelForSeq2SeqLM
from huggingface_hub import login

# Ler o token da variável de ambiente
hf_token = os.getenv("HUGGINGFACE_TOKEN")

if hf_token:
    login(token=hf_token)

# Carregar o modelo YOLOv5
model = torch.hub.load('ultralytics/yolov5', 'yolov5s')

# Função para calcular a GLCM e o contraste manualmente
def calculate_glcm_contrast(image):
    gray_image = cv2.cvtColor(np.array(image), cv2.COLOR_BGR2GRAY)
    max_value = gray_image.max() + 1
    glcm = np.zeros((max_value, max_value), dtype=np.float64)

    for i in range(gray_image.shape[0] - 1):
        for j in range(gray_image.shape[1] - 1):
            x = gray_image[i, j]
            y = gray_image[i + 1, j + 1]
            glcm[x, y] += 1

    glcm = glcm / glcm.sum()

    contrast = 0.0
    for i in range(max_value):
        for j in range(max_value):
            contrast += (i - j) ** 2 * glcm[i, j]
    
    return contrast

# Função para descrever imagem usando BLIP
def describe_image(image):
    processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-base")
    model = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-base")
    inputs = processor(image, return_tensors="pt")
    out = model.generate(**inputs)
    description = processor.decode(out[0], skip_special_tokens=True)
    return description

# Função para traduzir descrição para português
def translate_description(description):
    model_name = 'Helsinki-NLP/opus-mt-tc-big-en-pt'
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
    translated = model.generate(**tokenizer(description, return_tensors="pt", padding=True))
    translated_text = tokenizer.decode(translated[0], skip_special_tokens=True)
    return translated_text

# Função principal para processar imagem e gerar saída de voz
def process_image(image):
    # Detecção de objetos
    results = model(image)
    detected_image = results.render()[0]

    # Análise de cor (média RGB)
    mean_rgb = np.mean(np.array(image), axis=(0, 1))

    # Análise de textura
    texture_contrast = calculate_glcm_contrast(image)

    # Descrição da imagem
    description = describe_image(image)
    translated_description = translate_description(description)

    # Texto para voz
    tts = gTTS(text=translated_description, lang='pt')
    attempts = 0
    while attempts < 5:
        try:
            tts.save("output.mp3")
            break
        except gTTS.tts.gTTSError as e:
            if e.r.status_code == 429:
                print("Too many requests. Waiting before retrying...")
                time.sleep(5)
                attempts += 1
            else:
                raise e

    # Retornar imagem com detecções, descrição e áudio
    return Image.fromarray(detected_image), translated_description, "output.mp3"

# Carregar imagem de exemplo diretamente do código
example_image_path = "example1.JPG"

# Interface Gradio
iface = gr.Interface(
    fn=process_image,
    inputs=gr.Image(type="pil"),
    outputs=[gr.Image(type="pil"), gr.Textbox(), gr.Audio(type="filepath")],
    examples=[example_image_path]
)

iface.launch()