annotator/zoe/zoedepth/utils/geometry.py

# MIT License

# Copyright (c) 2022 Intelligent Systems Lab Org

# Permission is hereby granted, free of charge, to any person obtaining a copy
# of this software and associated documentation files (the "Software"), to deal
# in the Software without restriction, including without limitation the rights
# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
# copies of the Software, and to permit persons to whom the Software is
# furnished to do so, subject to the following conditions:

# The above copyright notice and this permission notice shall be included in all
# copies or substantial portions of the Software.

# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
# SOFTWARE.

# File author: Shariq Farooq Bhat

import numpy as np

def get_intrinsics(H,W):
    """
    Intrinsics for a pinhole camera model.
    Assume fov of 55 degrees and central principal point.
    """
    f = 0.5 * W / np.tan(0.5 * 55 * np.pi / 180.0)
    cx = 0.5 * W
    cy = 0.5 * H
    return np.array([[f, 0, cx],
                     [0, f, cy],
                     [0, 0, 1]])

def depth_to_points(depth, R=None, t=None):

    K = get_intrinsics(depth.shape[1], depth.shape[2])
    Kinv = np.linalg.inv(K)
    if R is None:
        R = np.eye(3)
    if t is None:
        t = np.zeros(3)

    # M converts from your coordinate to PyTorch3D's coordinate system
    M = np.eye(3)
    M[0, 0] = -1.0
    M[1, 1] = -1.0

    height, width = depth.shape[1:3]

    x = np.arange(width)
    y = np.arange(height)
    coord = np.stack(np.meshgrid(x, y), -1)
    coord = np.concatenate((coord, np.ones_like(coord)[:, :, [0]]), -1)  # z=1
    coord = coord.astype(np.float32)
    # coord = torch.as_tensor(coord, dtype=torch.float32, device=device)
    coord = coord[None]  # bs, h, w, 3

    D = depth[:, :, :, None, None]
    # print(D.shape, Kinv[None, None, None, ...].shape, coord[:, :, :, :, None].shape )
    pts3D_1 = D * Kinv[None, None, None, ...] @ coord[:, :, :, :, None]
    # pts3D_1 live in your coordinate system. Convert them to Py3D's
    pts3D_1 = M[None, None, None, ...] @ pts3D_1
    # from reference to targe tviewpoint
    pts3D_2 = R[None, None, None, ...] @ pts3D_1 + t[None, None, None, :, None]
    # pts3D_2 = pts3D_1
    # depth_2 = pts3D_2[:, :, :, 2, :]  # b,1,h,w
    return pts3D_2[:, :, :, :3, 0][0]


def create_triangles(h, w, mask=None):
    """
    Reference: https://github.com/google-research/google-research/blob/e96197de06613f1b027d20328e06d69829fa5a89/infinite_nature/render_utils.py#L68
    Creates mesh triangle indices from a given pixel grid size.
        This function is not and need not be differentiable as triangle indices are
        fixed.
    Args:
    h: (int) denoting the height of the image.
    w: (int) denoting the width of the image.
    Returns:
    triangles: 2D numpy array of indices (int) with shape (2(W-1)(H-1) x 3)
    """
    x, y = np.meshgrid(range(w - 1), range(h - 1))
    tl = y * w + x
    tr = y * w + x + 1
    bl = (y + 1) * w + x
    br = (y + 1) * w + x + 1
    triangles = np.array([tl, bl, tr, br, tr, bl])
    triangles = np.transpose(triangles, (1, 2, 0)).reshape(
        ((w - 1) * (h - 1) * 2, 3))
    if mask is not None:
        mask = mask.reshape(-1)
        triangles = triangles[mask[triangles].all(1)]
    return triangles