def __getitem__(self, index):
img_path = self.filelist[index]
img = convert_to_rgb(io.imread(img_path))
img = resize_max_length(img, 240)
img = img[:, :, 0].astype(np.float32)
img /= 255.0
return img
def __getitem__(self, index):
img_path = self.img_paths[index]
img = io.imread(img_path)
img = convert_to_rgb(img)
img0 = img
# h, w = (240, 240)
# img0 = random_crop(img, size=(h, w))
img1, H, scale = homography_adaption(img0)
if np.random.uniform(0, 1) > 0.5:
# swap two images
img = img0
img0 = img1
img1 = img
H = np.linalg.inv(H)
h, w = img1.shape[:2]
scale = compute_scale(np.linalg.inv(H), h, w)
scale = 1. / scale
# the pixel scale change of img0 to img1 for each pixel in img0
h, w = img0.shape[:2]
left_scale = compute_scale(H, h, w)
left_scale = 1. / left_scale
pix_pos0, pix_pos1 = sample_ground_truth(img0, H)
# img = draw_corspd_region(img0, img1, H)
# import matplotlib.pyplot as plt
# plt.imshow(np.concatenate([np.concatenate([img, draw_scale(scale)], axis=1), draw_scale(left_scale)], axis=1))
# plt.show()
# img = draw_kps(img0, pix_pos0, img1, pix_pos1)
# import matplotlib.pyplot as plt
# _, (ax1, ax2) = plt.subplots(1, 2)
# ax1.imshow(img)
# ax2.imshow(msk)
# plt.show()
if self.transforms is not None:
img0, pix_pos0 = self.transforms(img0, pix_pos0)
img1, pix_pos1 = self.transforms(img1, pix_pos1)
pix_pos2 = sample_negative(img1, pix_pos1)
img0 = torch.tensor(img0).permute(2, 0, 1).float()
img1 = torch.tensor(img1).permute(2, 0, 1).float()
pix_pos0 = torch.tensor(pix_pos0).float()
pix_pos1 = torch.tensor(pix_pos1).float()
target = dict(kps0=pix_pos0,
kps1=pix_pos1,
kps2=pix_pos2,
H=H,
scale=scale,
left_scale=left_scale)
return img0, img1, target, index
def __getitem__(self, index):
img_path = self.img_paths[index]
# H_path = self.H_paths[index]
H_path = np.random.choice(self.H_paths)
img = io.imread(img_path)
img = convert_to_rgb(img)
# img0 = img
h, w = (240, 240)
img0 = cv2.resize(img.copy(), (w, h), interpolation=cv2.INTER_LINEAR)
# img0 = random_crop(img, size=(240, 240))
H = read_H(H_path, img0)
img1, scale = homography_adaption(img0, H)
# the pixel scale change of img0 to img1 for each pixel in img0
h, w = img0.shape[:2]
left_scale = compute_scale(H, h, w)
left_scale = 1. / left_scale
pix_pos0, pix_pos1 = sample_ground_truth(img0, H)
_, msk = get_homography_correspondence(h, w, np.linalg.inv(H))
msk = msk.astype(np.float32)
# img = draw_corspd_region(img0, img1, H)
# import matplotlib.pyplot as plt
# plt.imshow(np.concatenate([img, draw_scale(scale)], axis=1))
# plt.show()
# img = draw_kps(img0, pix_pos0, img1, pix_pos1)
# import matplotlib.pyplot as plt
# plt.imshow(np.concatenate([img0, img], axis=1))
# plt.show()
if self.transforms is not None:
img0, pix_pos0 = self.transforms(img0, pix_pos0)
img1, pix_pos1 = self.transforms(img1, pix_pos1)
pix_pos2 = sample_negative(img1, pix_pos1)
img0 = torch.tensor(img0).permute(2, 0, 1).float()
img1 = torch.tensor(img1).permute(2, 0, 1).float()
pix_pos0 = torch.tensor(pix_pos0).float()
pix_pos1 = torch.tensor(pix_pos1).float()
scale = torch.tensor(scale).unsqueeze(0)
left_scale = torch.tensor(left_scale).unsqueeze(0)
msk = torch.tensor(msk).unsqueeze(0)
target = dict(kps0=pix_pos0,
kps1=pix_pos1,
kps2=pix_pos2,
H=H,
scale=scale,
msk=msk,
left_scale=left_scale)
return img0, img1, target, index
def __getitem__(self, index):
img_path = self.img_paths[index]
img = io.imread(img_path)
img = convert_to_rgb(img)
h, w = (240, 240)
img0 = cv2.resize(img.copy(), (w, h), interpolation=cv2.INTER_LINEAR)
# img0 = random_crop(img, size=(h, w))
img1, H, scale = homography_adaption(img0)
if np.random.uniform(0, 1) > 0.5:
# swap two images
img = img0
img0 = img1
img1 = img
H = np.linalg.inv(H)
h, w = img1.shape[:2]
scale = compute_scale(np.linalg.inv(H), h, w)
scale = 1. / scale
pix_pos0, pix_pos1 = sample_ground_truth(img0, H)
_, msk = get_homography_correspondence(h, w, np.linalg.inv(H))
msk = msk.astype(np.float32)
# img = draw_corspd_region(img0, img1, H)
# import matplotlib.pyplot as plt
# print(scale)
# plt.imshow(np.concatenate([img, draw_scale(scale)], axis=1))
# plt.show()
# img = draw_kps(img0, pix_pos0, img1, pix_pos1)
# import matplotlib.pyplot as plt
# _, (ax1, ax2) = plt.subplots(1, 2)
# ax1.imshow(img)
# ax2.imshow(msk)
# plt.show()
if self.transforms is not None:
img0, pix_pos0 = self.transforms(img0, pix_pos0)
img1, pix_pos1 = self.transforms(img1, pix_pos1)
pix_pos2 = sample_negative(img1, pix_pos1)
img0 = torch.tensor(img0).permute(2, 0, 1).float()
img1 = torch.tensor(img1).permute(2, 0, 1).float()
pix_pos0 = torch.tensor(pix_pos0).float()
pix_pos1 = torch.tensor(pix_pos1).float()
target = dict(kps0=pix_pos0,
kps1=pix_pos1,
kps2=pix_pos2,
H=H,
scale=scale,
msk=msk)
return img0, img1, target, index
def read_img(img_path, scale):
img = convert_to_rgb(io.imread(img_path))
oh, ow = img.shape[:2]
if isinstance(scale, int):
img = resize_max_length(img, scale)
elif isinstance(scale, tuple):
img = cv2.resize(img, scale)
h, w = img.shape[:2]
scale_h, scale_w = h / oh, w / ow
return img, scale_h, scale_w
def __getitem__(self, index):
img_path = self.img_paths[index]
img = io.imread(img_path)
img = convert_to_rgb(img)
img0 = img
# img0 = random_crop(img, size=(240, 320))
img1, H, patch_ratio = homography_adaption(img0, angle=self.angle)
pix_pos0, pix_pos1 = sample_ground_truth(img0, H)
# img = draw_corspd_region(img0, img1, H)
# import matplotlib.pyplot as plt
# plt.imshow(img)
# plt.show()
if self.transforms is not None:
img0, pix_pos0 = self.transforms(img0, pix_pos0)
img1, pix_pos1 = self.transforms(img1, pix_pos1)
pix_pos2 = sample_negative(img1, pix_pos1)
img0 = torch.tensor(img0).permute(2, 0, 1).float()
img1 = torch.tensor(img1).permute(2, 0, 1).float()
# img = draw_triplet(img0, pix_pos0, img1, pix_pos1, img1, pix_pos2)
# import matplotlib.pyplot as plt
# plt.imshow(img)
# plt.show()
# return
if patch_ratio > 1. / np.sqrt(2):
scale = 1
else:
scale = 2
pix_pos0 = torch.tensor(pix_pos0).float()
pix_pos1 = torch.tensor(pix_pos1).float()
target = dict(kps0=pix_pos0,
kps1=pix_pos1,
kps2=pix_pos2,
H=H,
scale=scale)
return img0, img1, target, index
def __getitem__(self, index):
img_path = self.img_paths[index]
img = io.imread(img_path)
img = convert_to_rgb(img)
img0 = img
# let p, p' be corresponding points in the two images, then
# area[p'] / area[p] = scale[p']
# img0 = random_crop(img, size=(240, 320))
img1, H, scale = homography_adaption(img0)
pix_pos0, pix_pos1 = sample_ground_truth(img0, H)
if self.transforms is not None:
img0, pix_pos0 = self.transforms(img0, pix_pos0)
img1, pix_pos1 = self.transforms(img1, pix_pos1)
pix_pos2 = sample_negative(img1, pix_pos1)
img0 = torch.tensor(img0).permute(2, 0, 1).float()
img1 = torch.tensor(img1).permute(2, 0, 1).float()
# img = draw_triplet(img0, pix_pos0, img1, pix_pos1, img1, pix_pos2)
# import matplotlib.pyplot as plt
# plt.imshow(img)
# plt.show()
# return
pix_pos0 = torch.tensor(pix_pos0).float()
pix_pos1 = torch.tensor(pix_pos1).float()
# switch image so that the second is smaller
h, w = img1.size()[1:]
img0, img1 = img1, img0
pix_pos0, pix_pos1 = pix_pos1, pix_pos0
H = np.linalg.inv(H)
scale = 1.0 / compute_scale(np.linalg.inv(H), h, w)
target = dict(kps0=pix_pos0,
kps1=pix_pos1,
kps2=pix_pos2,
H=H,
scale=scale)
return img0, img1, target, index
def __getitem__(self, index):
img_path = self.img_paths[index]
img = io.imread(img_path)
img = convert_to_rgb(img)
# img0 = img
img0 = resize_max_length(img, 640)
img1, H = homography_adaption(img0, angle=self.angle)
pix_pos0, pix_pos1 = sample_ground_truth(img0, H)
# img = draw_corspd_region(img0, img1, H)
# import matplotlib.pyplot as plt
# plt.imshow(img)
# plt.show()
if self.transforms is not None:
img0, pix_pos0 = self.transforms(img0, pix_pos0)
img1, pix_pos1 = self.transforms(img1, pix_pos1)
pix_pos2 = sample_negative(img1, pix_pos1)
img0 = torch.tensor(img0).permute(2, 0, 1).float()
img1 = torch.tensor(img1).permute(2, 0, 1).float()
# img = draw_triplet(img0, pix_pos0, img1, pix_pos1, img1, pix_pos2)
# import matplotlib.pyplot as plt
# plt.imshow(img)
# plt.show()
# return
pix_pos0 = torch.tensor(pix_pos0).float()
pix_pos1 = torch.tensor(pix_pos1).float()
target = dict(
kps0=pix_pos0,
kps1=pix_pos1,
kps2=pix_pos2,
H=H,
)
return img0, img1, target, index