SegCloth.py

from transformers import pipeline
from PIL import Image, ImageChops, ImageOps
import numpy as np
from io import BytesIO
import base64
from transparent_background import Remover

# Initialisation du pipeline de segmentation
segmenter = pipeline(model="mattmdjaga/segformer_b2_clothes")

#@spaces.GPU
def remove_background(image):
    remover = Remover()
    if isinstance(image, Image.Image):
        output = remover.process(image)
    elif isinstance(image, np.ndarray):
        image_pil = Image.fromarray(image)
        output = remover.process(image_pil)
    else:
        raise TypeError("Unsupported image type")
    return output
    
def encode_image_to_base64(image):
    buffered = BytesIO()
    image.save(buffered, format="PNG")
    return base64.b64encode(buffered.getvalue()).decode('utf-8')

def segment_clothing(img, clothes=["Hat", "Upper-clothes", "Skirt", "Pants", "Dress", "Belt", "Left-shoe", "Right-shoe", "Scarf"], margin=10):
    # Segmentation de l'image
    segments = segmenter(img)
    
    # Liste des images segmentées
    result_images = []

    for s in segments:
        if s['label'] in clothes:
            # Conversion du masque en tableau NumPy
            mask_array = np.array(s['mask'])
            
            # Création d'une image vide avec transparence
            empty_image = Image.new("RGBA", img.size, (0, 0, 0, 0))
            
            # Conversion du masque en image PIL (niveau de gris)
            mask_image = Image.fromarray(mask_array).convert("L")
            
            # Extraction de la partie de l'image correspondant au masque
            segmented_part = ImageChops.multiply(img.convert("RGBA"), Image.merge("RGBA", [mask_image, mask_image, mask_image, mask_image]))

            # Application du masque sur l'image vide
            empty_image.paste(segmented_part, mask=mask_image)
            
            # Déterminer la bounding box du masque
            bbox = mask_image.getbbox()
            if bbox:
                # Ajouter la marge autour de la bounding box
                left, top, right, bottom = bbox
                left = max(0, left - margin)
                top = max(0, top - margin)
                right = min(img.width, right + margin)
                bottom = min(img.height, bottom + margin)
                
                # Recadrer l'image à la taille du masque avec la marge
                cropped_image = empty_image.crop((left, top, right, bottom))
                
                # Encodage de l'image recadrée en base64
                image_rm_background = remove_background(cropped_image)
                imageBase64 = encode_image_to_base64(image_rm_background)
                #result_images.append((s['label'], imageBase64))
                result_images.append(imageBase64)

    return result_images