src/run_deeplab.py

# Copyright (c) 2020, Roy Or-El. All rights reserved.
#
# This work is licensed under the Creative Commons
# Attribution-NonCommercial-ShareAlike 4.0 International License.
# To view a copy of this license, visit
# http://creativecommons.org/licenses/by-nc-sa/4.0/ or send a letter to
# Creative Commons, PO Box 1866, Mountain View, CA 94042, USA.

# This code is a modification of the main.py file
# from the https://github.com/chenxi116/DeepLabv3.pytorch repository

import os
import argparse
import requests
import numpy as np
from tqdm import tqdm
from pathlib import Path

import torch
from PIL import Image

from src.models import deeplab
from src.data.data_loader import CelebASegmentation
from src.utils.deeplab_util import download_file


# python -m src.run_deeplab --resolution 512 --dataset_root ./images/images512x512 --output_dir ./segmaps/segmaps512x512

parser = argparse.ArgumentParser()
parser.add_argument('--resolution', type=int, default=256,
					help='segmentation output size')
parser.add_argument('--dataset_root', type=str, default="./images/images512x512",
					help='Path to images dir')
parser.add_argument('--output_dir', type=str, default="./segmaps/segmaps512x512",
					help='Path to output segmap dir')
# parser.add_argument('--workers', type=int, default=4,
# 					help='number of data loading workers')
args = parser.parse_args()


resnet_file_spec = dict(file_url='https://drive.google.com/uc?id=1oRGgrI4KNdefbWVpw0rRkEP1gbJIRokM', file_path='deeplab_model/R-101-GN-WS.pth.tar', file_size=178260167, file_md5='aa48cc3d3ba3b7ac357c1489b169eb32')
deeplab_file_spec = dict(file_url='https://drive.google.com/uc?id=1w2XjDywFr2NjuUWaLQDRktH7VwIfuNlY', file_path='deeplab_model/deeplab_model.pth', file_size=464446305, file_md5='8e8345b1b9d95e02780f9bed76cc0293')

def main():
    resolution = args.resolution
    assert torch.cuda.is_available()
    torch.backends.cudnn.benchmark = True
    model_fname = 'deeplab_model/deeplab_model.pth'
    # dataset_root = 'ffhq_aging{}x{}'.format(resolution,resolution)
    assert os.path.isdir(args.dataset_root)
    # dataset = CelebASegmentation(args.dataset_root)
    dataset = CelebASegmentation(args.dataset_root, crop_size=513) # ~!

    if not os.path.isfile(resnet_file_spec['file_path']):
        print('Downloading backbone Resnet Model parameters')
        with requests.Session() as session:
            download_file(session, resnet_file_spec)

        print('Done!')

    model = getattr(deeplab, 'resnet101')(
    	      pretrained=True,
    	      num_classes=len(dataset.CLASSES),
    	      num_groups=32,
    	      weight_std=True,
    	      beta=False)

    model = model.cuda()
    model.eval()
    if not os.path.isfile(deeplab_file_spec['file_path']):
        print('Downloading DeeplabV3 Model parameters')
        with requests.Session() as session:
            download_file(session, deeplab_file_spec)

        print('Done!')

    checkpoint = torch.load(model_fname)
    state_dict = {k[7:]: v for k, v in checkpoint['state_dict'].items() if 'tracked' not in k}
    model.load_state_dict(state_dict)
    
    for i in tqdm(range(len(dataset))):
        inputs=dataset[i]
        inputs = inputs.cuda()
        outputs = model(inputs.unsqueeze(0))
        _, pred = torch.max(outputs, 1)
        pred = pred.data.cpu().numpy().squeeze().astype(np.uint8)
        mask_pred = Image.fromarray(pred)
        
        mask_pred=mask_pred.resize((resolution,resolution), Image.NEAREST)
        
        if not os.path.exists(args.output_dir):
            os.makedirs(args.output_dir)
        
        savename = os.path.join(args.output_dir, f"{Path(dataset.images[i]).stem}.png")
        mask_pred.save(savename)
        
        # imname = os.path.basename(dataset.images[i])
        # try:
        #     mask_pred.save(dataset.images[i].replace(imname,'parsings/'+imname[:-4]+'.png'))
        # except FileNotFoundError:
        #     os.makedirs(os.path.join(os.path.dirname(dataset.images[i]),'parsings'))
        #     mask_pred.save(dataset.images[i].replace(imname,'parsings/'+imname[:-4]+'.png'))

        print('processed {0}/{1} images'.format(i + 1, len(dataset)))

if __name__ == "__main__":
  main()