def __getitem__(self, index):
rgb, sparse, target, rgb_near = self.__getraw__(index)
rgb, sparse, target, rgb_near = self.transform(rgb, sparse, target,
rgb_near, self.args)
r_mat, t_vec = None, None
if self.split == 'train' and self.args.use_pose:
success, r_vec, t_vec = get_pose_pnp(rgb, rgb_near, sparse, self.K)
# discard if translation is too small
success = success and LA.norm(t_vec) > self.threshold_translation
if success:
r_mat, _ = cv2.Rodrigues(r_vec)
else:
# return the same image and no motion when PnP fails
rgb_near = rgb
t_vec = np.zeros((3, 1))
r_mat = np.eye(3)
rgb, gray = handle_gray(rgb, self.args)
candidates = {"rgb":rgb, "d":sparse, "gt":target, \
"g":gray, "r_mat":r_mat, "t_vec":t_vec, "rgb_near":rgb_near}
items = {
key: to_float_tensor(val)
for key, val in candidates.items() if val is not None
}
return items
def __getitem__(self, index):
# target intensity as intensity_real
rgb, sparse, target, rgb_near, sparse_intensity, target_intensity, target_intensity_pure = self.__getraw__(
index)
rgb, sparse, target, rgb_near, sparse_intensity, target_intensity, target_intensity_pure \
= self.transform(rgb, sparse, target,sparse_intensity, target_intensity, rgb_near, self.args, target_intensity_pure)
r_mat, t_vec = None, None
if self.split == 'train' and self.args.use_pose:
success, r_vec, t_vec = get_pose_pnp(rgb, rgb_near, sparse, self.K)
# discard if translation is too small
success = success and LA.norm(t_vec) > self.threshold_translation
if success:
r_mat, _ = cv2.Rodrigues(r_vec)
else:
# return the same image and no motion when PnP fails
rgb_near = rgb
t_vec = np.zeros((3, 1))
r_mat = np.eye(3)
rgb, gray = handle_gray(rgb, self.args)
# Ireal as gt_intensity
candidates = {"rgb":rgb, "d":sparse, "gt":target, \
"intensity":sparse_intensity, "gt_intensity":target_intensity, \
"g":gray, "r_mat":r_mat, "t_vec":t_vec, "rgb_near":rgb_near, \
"gt_intensity_pure":target_intensity_pure}
items = {
key: to_float_tensor(val)
for key, val in candidates.items() if val is not None
}
filename = self.paths['d'][index].split('/')[-1]
items["filename"] = filename
return items
def __getitem__(self, index):
rgb, sparse, target, rgb_near = self.__getraw__(index)
#im = Image.fromarray(rgb)
#im.save("orig.jpg")
print(f"sparse: {len(np.where(sparse)[0] > 0)}")
print(f"dense gt: {len(np.where(target)[0] > 0)}")
rgb, sparse, target, rgb_near = self.transform(rgb, sparse, target,
rgb_near, self.args)
#im = Image.fromarray(rgb)
#im.save("orig_trans.jpg")
print("\ntransform")
print(f"sparse: {len(np.where(sparse)[0] > 0)}")
print(f"dense gt: {len(np.where(target)[0] > 0)}")
r_mat, t_vec = None, None
if self.split == 'train' and self.args.use_pose:
success, r_vec, t_vec = get_pose_pnp(rgb, rgb_near, sparse, self.K)
# discard if translation is too small
success = success and LA.norm(t_vec) > self.threshold_translation
if success:
r_mat, _ = cv2.Rodrigues(r_vec)
else:
# return the same image and no motion when PnP fails
rgb_near = rgb
t_vec = np.zeros((3, 1))
r_mat = np.eye(3)
rgb_col = rgb
#return rgb None if the option in args.input is not rgb
rgb, gray = handle_gray(rgb, self.args)
candidates = {"rgb":rgb, "d":sparse, "gt":target, \
"g":gray, "r_mat":r_mat, "t_vec":t_vec, "rgb_near":rgb_near}
items = {
key: to_float_tensor(val)
for key, val in candidates.items() if val is not None
}
items['d_path'] = self.paths['d'][index]
return items