search

milmillin / ReparamCAD

15 stars 2 forks source link

the Rendered Problem #3

Closed liyueying233 closed 7 months ago

liyueying233 commented 8 months ago

Hi！I met some difficulties about generating the rendered pictures from mesh, which have almost one kind of color.

image

I can not solve this problem so that my output is totally wrong. Could you please help me solving this problem?

milmillin commented 8 months ago

Hi,

Apologies. Please change the projection function in rpcad/camera.py to the following:

def projection(fov_rad: float = np.radians(45.0), ar=1.0, near=1.0, far=50.0):
    """
    From https://github.com/rgl-epfl/large-steps-pytorch by @bathal1 (Baptiste Nicolet)

    Build a perspective projection matrix.
    Parameters
    ----------
    fov_x : float
        Horizontal field of view (in degrees).
    ar : float
        Aspect ratio (w/h).
    near : float
        Depth of the near plane relative to the camera.
    far : float
        Depth of the far plane relative to the camera.
    """
    tanhalffov = np.tan((fov_rad / 2))
    max_y = tanhalffov * near
    min_y = -max_y
    max_x = max_y * ar
    min_x = -max_x

    z_sign = -1.0
    proj_mat = np.array([[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]])

    proj_mat[0, 0] = 2.0 * near / (max_x - min_x)
    proj_mat[1, 1] = 2.0 * near / (max_y - min_y)
    proj_mat[0, 2] = (max_x + min_x) / (max_x - min_x)
    proj_mat[1, 2] = (max_y + min_y) / (max_y - min_y)
    proj_mat[3, 2] = z_sign

    proj_mat[2, 2] = z_sign * far / (far - near)
    proj_mat[2, 3] = -(far * near) / (far - near)

    return proj_mat

It was a mix of radians and degrees. I might have pulled out the wrong version of the code. I am looking to see further inaccuracies before I push the fixes.

liyueying233 commented 7 months ago

Hi,

Apologies. Please change the projection function in rpcad/camera.py to the following:

def projection(fov_rad: float = np.radians(45.0), ar=1.0, near=1.0, far=50.0):
    """
    From https://github.com/rgl-epfl/large-steps-pytorch by @bathal1 (Baptiste Nicolet)

    Build a perspective projection matrix.
    Parameters
    ----------
    fov_x : float
        Horizontal field of view (in degrees).
    ar : float
        Aspect ratio (w/h).
    near : float
        Depth of the near plane relative to the camera.
    far : float
        Depth of the far plane relative to the camera.
    """
    tanhalffov = np.tan((fov_rad / 2))
    max_y = tanhalffov * near
    min_y = -max_y
    max_x = max_y * ar
    min_x = -max_x

    z_sign = -1.0
    proj_mat = np.array([[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 0, 0]])

    proj_mat[0, 0] = 2.0 * near / (max_x - min_x)
    proj_mat[1, 1] = 2.0 * near / (max_y - min_y)
    proj_mat[0, 2] = (max_x + min_x) / (max_x - min_x)
    proj_mat[1, 2] = (max_y + min_y) / (max_y - min_y)
    proj_mat[3, 2] = z_sign

    proj_mat[2, 2] = z_sign * far / (far - near)
    proj_mat[2, 3] = -(far * near) / (far - near)

    return proj_mat

It was a mix of radians and degrees. I might have pulled out the wrong version of the code. I am looking to see further inaccuracies before I push the fixes.

image Sorry about forgeting to reply that the question is solved successfully. Thanks a lot for your help!