File size: 3,460 Bytes
7decfe1 | 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 | # Copyright 2023-2025 Marigold Team, ETH Zürich. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# --------------------------------------------------------------------------
# More information about Marigold:
# https://marigoldmonodepth.github.io
# https://marigoldcomputervision.github.io
# Efficient inference pipelines are now part of diffusers:
# https://huggingface.co/docs/diffusers/using-diffusers/marigold_usage
# https://huggingface.co/docs/diffusers/api/pipelines/marigold
# Examples of trained models and live demos:
# https://huggingface.co/prs-eth
# Related projects:
# https://rollingdepth.github.io/
# https://marigolddepthcompletion.github.io/
# Citation (BibTeX):
# https://github.com/prs-eth/Marigold#-citation
# If you find Marigold useful, we kindly ask you to cite our papers.
# --------------------------------------------------------------------------
import numpy as np
import torch
def align_depth_least_square(
gt_arr: np.ndarray,
pred_arr: np.ndarray,
valid_mask_arr: np.ndarray,
return_scale_shift=True,
max_resolution=None,
):
ori_shape = pred_arr.shape # input shape
gt = gt_arr.squeeze() # [H, W]
pred = pred_arr.squeeze()
valid_mask = valid_mask_arr.squeeze()
# Downsample
if max_resolution is not None:
scale_factor = np.min(max_resolution / np.array(ori_shape[-2:]))
if scale_factor < 1:
downscaler = torch.nn.Upsample(scale_factor=scale_factor, mode="nearest")
gt = downscaler(torch.as_tensor(gt).unsqueeze(0)).numpy()
pred = downscaler(torch.as_tensor(pred).unsqueeze(0)).numpy()
valid_mask = (
downscaler(torch.as_tensor(valid_mask).unsqueeze(0).float())
.bool()
.numpy()
)
assert (
gt.shape == pred.shape == valid_mask.shape
), f"{gt.shape}, {pred.shape}, {valid_mask.shape}"
gt_masked = gt[valid_mask].reshape((-1, 1))
pred_masked = pred[valid_mask].reshape((-1, 1))
# numpy solver
_ones = np.ones_like(pred_masked)
A = np.concatenate([pred_masked, _ones], axis=-1)
X = np.linalg.lstsq(A, gt_masked, rcond=None)[0]
scale, shift = X
aligned_pred = pred_arr * scale + shift
# restore dimensions
aligned_pred = aligned_pred.reshape(ori_shape)
if return_scale_shift:
return aligned_pred, scale, shift
else:
return aligned_pred
def depth2disparity(depth, return_mask=False):
if isinstance(depth, torch.Tensor):
disparity = torch.zeros_like(depth)
elif isinstance(depth, np.ndarray):
disparity = np.zeros_like(depth)
non_negtive_mask = depth > 0
disparity[non_negtive_mask] = 1.0 / depth[non_negtive_mask]
if return_mask:
return disparity, non_negtive_mask
else:
return disparity
def disparity2depth(disparity, **kwargs):
return depth2disparity(disparity, **kwargs)
|