File size: 5,296 Bytes
1b65314 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 |
# Copyright (C) 2024-present Naver Corporation. All rights reserved.
# Licensed under CC BY-NC-SA 4.0 (non-commercial use only).
#
# --------------------------------------------------------
# utilitary functions about images (loading/converting...)
# --------------------------------------------------------
import os
import torch
import numpy as np
import PIL.Image
from PIL.ImageOps import exif_transpose
import torchvision.transforms as tvf
os.environ["OPENCV_IO_ENABLE_OPENEXR"] = "1"
import cv2 # noqa
try:
from pillow_heif import register_heif_opener # noqa
register_heif_opener()
heif_support_enabled = True
except ImportError:
heif_support_enabled = False
ImgNorm = tvf.Compose([tvf.ToTensor(), tvf.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
def img_to_arr( img ):
if isinstance(img, str):
img = imread_cv2(img)
return img
def imread_cv2(path, options=cv2.IMREAD_COLOR):
""" Open an image or a depthmap with opencv-python.
"""
if path.endswith(('.exr', 'EXR')):
options = cv2.IMREAD_ANYDEPTH
img = cv2.imread(path, options)
if img is None:
raise IOError(f'Could not load image={path} with {options=}')
if img.ndim == 3:
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
return img
def rgb(ftensor, true_shape=None):
if isinstance(ftensor, list):
return [rgb(x, true_shape=true_shape) for x in ftensor]
if isinstance(ftensor, torch.Tensor):
ftensor = ftensor.detach().cpu().numpy() # H,W,3
if ftensor.ndim == 3 and ftensor.shape[0] == 3:
ftensor = ftensor.transpose(1, 2, 0)
elif ftensor.ndim == 4 and ftensor.shape[1] == 3:
ftensor = ftensor.transpose(0, 2, 3, 1)
if true_shape is not None:
H, W = true_shape
ftensor = ftensor[:H, :W]
if ftensor.dtype == np.uint8:
img = np.float32(ftensor) / 255
else:
img = (ftensor * 0.5) + 0.5
return img.clip(min=0, max=1)
def _resize_pil_image(img, long_edge_size):
S = max(img.size)
if S > long_edge_size:
interp = PIL.Image.LANCZOS
elif S <= long_edge_size:
interp = PIL.Image.BICUBIC
new_size = tuple(int(round(x*long_edge_size/S)) for x in img.size)
return img.resize(new_size, interp)
def load_images(folder_or_list, cog_seg_maps, size, square_ok=False, verbose=True):
""" open and convert all images in a list or folder to proper input format for DUSt3R
"""
if isinstance(folder_or_list, str):
if verbose:
print(f'>> Loading images from {folder_or_list}')
root, folder_content = folder_or_list, sorted(os.listdir(folder_or_list))
elif isinstance(folder_or_list, list):
if verbose:
print(f'>> Loading a list of {len(folder_or_list)} images')
root, folder_content = '', folder_or_list
else:
raise ValueError(f'bad {folder_or_list=} ({type(folder_or_list)})')
supported_images_extensions = ['.jpg', '.jpeg', '.png']
if heif_support_enabled:
supported_images_extensions += ['.heic', '.heif']
supported_images_extensions = tuple(supported_images_extensions)
imgs = []
for i, path in enumerate(folder_content):
if not path.lower().endswith(supported_images_extensions):
continue
img = exif_transpose(PIL.Image.open(os.path.join(root, path))).convert('RGB')
img_np = np.array(img)
smoothed_image = np.zeros_like(img_np)
seg_map = cog_seg_maps[i]
unique_labels = np.unique(seg_map)
for label in unique_labels:
mask = (seg_map == label)
mean_color = img_np[mask].mean(axis=0)
smoothed_image[mask] = mean_color
smoothed_image = cv2.addWeighted(img_np, 0.05, smoothed_image, 0.95, 0)
smoothed_image = PIL.Image.fromarray(smoothed_image)
W1, H1 = img.size
if size == 224:
# resize short side to 224 (then crop)
img = _resize_pil_image(img, round(size * max(W1/H1, H1/W1)))
smoothed_image = _resize_pil_image(smoothed_image, round(size * max(W1/H1, H1/W1)))
else:
# resize long side to 512
img = _resize_pil_image(img, size)
smoothed_image = _resize_pil_image(smoothed_image, size)
W, H = img.size
cx, cy = W//2, H//2
if size == 224:
half = min(cx, cy)
img = img.crop((cx-half, cy-half, cx+half, cy+half))
smoothed_image = smoothed_image.crop((cx-half, cy-half, cx+half, cy+half))
else:
halfw, halfh = ((2*cx)//16)*8, ((2*cy)//16)*8
if not (square_ok) and W == H:
halfh = 3*halfw/4
img = img.crop((cx-halfw, cy-halfh, cx+halfw, cy+halfh))
smoothed_image = smoothed_image.crop((cx-halfw, cy-halfh, cx+halfw, cy+halfh))
W2, H2 = img.size
if verbose:
print(f' - adding {path} with resolution {W1}x{H1} --> {W2}x{H2}')
imgs.append(dict(img=ImgNorm(smoothed_image)[None], ori_img=ImgNorm(img)[None], true_shape=np.int32(
[img.size[::-1]]), idx=len(imgs), instance=str(len(imgs))))
assert imgs, 'no images foud at '+root
if verbose:
print(f' (Found {len(imgs)} images)')
return imgs
|