osanseviero/Neural_Image_Colorizer

1import PIL2import torch3import torch.nn as nn4import cv25from skimage.color import lab2rgb, rgb2lab, rgb2gray6from skimage import io7import matplotlib.pyplot as plt8import numpy as np9 10class ColorizationNet(nn.Module):11  def __init__(self, input_size=128):12    super(ColorizationNet, self).__init__()13 14    MIDLEVEL_FEATURE_SIZE = 12815    resnet=models.resnet18(pretrained=True)16    resnet.conv1.weight=nn.Parameter(resnet.conv1.weight.sum(dim=1).unsqueeze(1))17    18    self.midlevel_resnet =nn.Sequential(*list(resnet.children())[0:6])19 20    self.upsample = nn.Sequential(     21      nn.Conv2d(MIDLEVEL_FEATURE_SIZE, 128, kernel_size=3, stride=1, padding=1),22      nn.BatchNorm2d(128),23      nn.ReLU(),24      nn.Upsample(scale_factor=2),25      nn.Conv2d(128, 64, kernel_size=3, stride=1, padding=1),26      nn.BatchNorm2d(64),27      nn.ReLU(),28      nn.Conv2d(64, 64, kernel_size=3, stride=1, padding=1),29      nn.BatchNorm2d(64),30      nn.ReLU(),31      nn.Upsample(scale_factor=2),32      nn.Conv2d(64, 32, kernel_size=3, stride=1, padding=1),33      nn.BatchNorm2d(32),34      nn.ReLU(),35      nn.Conv2d(32, 2, kernel_size=3, stride=1, padding=1),36      nn.Upsample(scale_factor=2)37    )38 39  def forward(self, input):40 41    # Pass input through ResNet-gray to extract features42    midlevel_features = self.midlevel_resnet(input)43 44    # Upsample to get colors45    output = self.upsample(midlevel_features)46    return output47 48    49    50def show_output(grayscale_input, ab_input):51  '''Show/save rgb image from grayscale and ab channels52     Input save_path in the form {'grayscale': '/path/', 'colorized': '/path/'}'''53  color_image = torch.cat((grayscale_input, ab_input), 0).detach().numpy() # combine channels54  color_image = color_image.transpose((1, 2, 0))  # rescale for matplotlib55  color_image[:, :, 0:1] = color_image[:, :, 0:1] * 10056  color_image[:, :, 1:3] = color_image[:, :, 1:3] * 255 - 128   57  color_image = lab2rgb(color_image.astype(np.float64))58  grayscale_input = grayscale_input.squeeze().numpy()59  # plt.imshow(grayscale_input)60  # plt.imshow(color_image)61  return color_image62 63def colorize(img,print_img=True):64    # img=cv2.imread(img)65    img=cv2.resize(img,(224,224))66    grayscale_input= torch.Tensor(rgb2gray(img))67    ab_input=model(grayscale_input.unsqueeze(0).unsqueeze(0)).squeeze(0)68    predicted=show_output(grayscale_input.unsqueeze(0), ab_input)69    if print_img:70        plt.imshow(predicted)71    return predicted72 73# device=torch.device("cuda" if torch.cuda.is_available() else "cpu")74# torch.load with map_location=torch.device('cpu') 75model=torch.load("model-final.pth",map_location ='cpu')76 77 78import streamlit as st79st.title("Image Colorizer")80st.write('\n')81st.write('Find more info at: https://github.com/Pranav082001/Neural-Image-Colorizer or at https://medium.com/@pranav.kushare2001/colorize-your-black-and-white-photos-using-ai-4652a34e967.')82 83# Sidebar84st.sidebar.title("Upload Image")85file=st.sidebar.file_uploader("Please upload a Black and White image",type=["jpg","jpeg","png"])86 87if st.sidebar.button("Colorize image"):88    with st.spinner('Colorizing...'):89      file_bytes = np.asarray(bytearray(file.read()), dtype=np.uint8)90      opencv_image = cv2.imdecode(file_bytes, 1)91      im=colorize(opencv_image)92    st.text("Original")93    st.image(file)94    st.text("Colorized!!")95    st.image(im)96