| from PIL import Image |
| from tqdm import tqdm |
| import numpy as np |
| import argparse |
| import glob |
| import pdb |
| import os |
|
|
| parser = argparse.ArgumentParser(description="resize image using bicubic upsampling") |
| parser.add_argument("--in_dir", default='/mnt/session_space/home/lingzzha/EgoHOS/data/train/label_foreground_512', type=str) |
| parser.add_argument("--out_dir", default='/mnt/session_space/home/lingzzha/EgoHOS/data/train/label_foreground_512_pad', type=str) |
| parser.add_argument("--pad_info_dir", default='/mnt/session_space/home/lingzzha/EgoHOS/data/train/label_foreground_512_pad_info', type=str) |
| parser.add_argument("--size", default=512, type=int) |
| args = parser.parse_args() |
|
|
| os.makedirs(args.out_dir, exist_ok=True) |
| os.makedirs(args.pad_info_dir, exist_ok=True) |
|
|
| i = 0 |
| for img_file in tqdm(glob.glob(args.in_dir+'/*')): |
|
|
| fname = img_file.split('/')[-1] |
|
|
| img = np.array(Image.open(img_file)) |
| w, h = img.shape[0], img.shape[1] |
| |
| if len(img.shape) == 3: |
| img_pad = np.zeros((args.size, args.size, img.shape[2])) |
| else: |
| img_pad = np.zeros((args.size, args.size)) |
|
|
| pad_info = np.zeros((1,2)) |
|
|
| if w < h: |
| if len(img.shape) == 3: |
| img_pad[args.size - w:, :, :] = img |
| else: |
| img_pad[args.size - w:, :] = img |
| pad_info[0, 0] = 0 |
| pad_info[0, 1] = w |
| else: |
| if len(img.shape) == 3: |
| img_pad[:, :h, :] = img |
| else: |
| img_pad[:, :h] = img |
| pad_info[0, 0] = 1 |
| pad_info[0, 1] = h |
| |
| img_pad = Image.fromarray(img_pad.astype('uint8')) |
| img_pad.save(os.path.join(args.out_dir, fname)) |
| |
| np.save(os.path.join(args.pad_info_dir, str(fname.split('.')[0])+'.npy'), pad_info) |
|
|
| i += 1 |
|
|
| |
|
|
|
|
|
|
| |
