File size: 3,702 Bytes
1e45055 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 |
import numpy as np
import torch
from common.utils import wrap
from common.quaternion import qrot, qinverse
def normalize_screen_coordinates(X, w, h):
assert X.shape[-1] == 2
# Normalize so that [0, w] is mapped to [-1, 1], while preserving the aspect ratio
return X/w*2 - [1, h/w]
def image_coordinates(X, w, h):
assert X.shape[-1] == 2
# Reverse camera frame normalization
return (X + [1, h/w])*w/2
def world_to_camera(X, R, t):
Rt = wrap(qinverse, R) # Invert rotation
return wrap(qrot, np.tile(Rt, (*X.shape[:-1], 1)), X - t) # Rotate and translate
def camera_to_world(X, R, t):
return wrap(qrot, np.tile(R, (*X.shape[:-1], 1)), X) + t
def project_to_2d(X, camera_params):
"""
Project 3D points to 2D using the Human3.6M camera projection function.
This is a differentiable and batched reimplementation of the original MATLAB script.
Arguments:
X -- 3D points in *camera space* to transform (N, *, 3)
camera_params -- intrinsic parameteres (N, 2+2+3+2=9)
"""
assert X.shape[-1] == 3
assert len(camera_params.shape) == 2
assert camera_params.shape[-1] == 9
assert X.shape[0] == camera_params.shape[0]
while len(camera_params.shape) < len(X.shape):
camera_params = camera_params.unsqueeze(1)
f = camera_params[..., :2]
c = camera_params[..., 2:4]
k = camera_params[..., 4:7]
p = camera_params[..., 7:]
XX = torch.clamp(X[..., :2] / X[..., 2:], min=-1, max=1)
r2 = torch.sum(XX[..., :2]**2, dim=len(XX.shape)-1, keepdim=True)
radial = 1 + torch.sum(k * torch.cat((r2, r2**2, r2**3), dim=len(r2.shape)-1), dim=len(r2.shape)-1, keepdim=True)
tan = torch.sum(p*XX, dim=len(XX.shape)-1, keepdim=True)
XXX = XX*(radial + tan) + p*r2
return f*XXX + c
def project_to_2d_linear(X, camera_params):
"""
Project 3D points to 2D using only linear parameters (focal length and principal point).
Arguments:
X -- 3D points in *camera space* to transform (N, *, 3)
camera_params -- intrinsic parameteres (N, 2+2+3+2=9)
"""
assert X.shape[-1] == 3
assert len(camera_params.shape) == 2
assert camera_params.shape[-1] == 9
assert X.shape[0] == camera_params.shape[0]
while len(camera_params.shape) < len(X.shape):
camera_params = camera_params.unsqueeze(1)
f = camera_params[..., :2]
c = camera_params[..., 2:4]
XX = torch.clamp(X[..., :2] / X[..., 2:], min=-1, max=1)
return f*XX + c
def uvd2xyz(uvd, gt_3D, cam):
"""
transfer uvd to xyz
:param uvd: N*T*V*3 (uv and z channel)
:param gt_3D: N*T*V*3 (NOTE: V=0 is absolute depth value of root joint)
:return: root-relative xyz results
"""
N, T, V,_ = uvd.size()
dec_out_all = uvd.view(-1, T, V, 3).clone() # N*T*V*3
root = gt_3D[:, :, 0, :].unsqueeze(-2).repeat(1, 1, V, 1).clone()# N*T*V*3
enc_in_all = uvd[:, :, :, :2].view(-1, T, V, 2).clone() # N*T*V*2
cam_f_all = cam[..., :2].view(-1,1,1,2).repeat(1,T,V,1) # N*T*V*2
cam_c_all = cam[..., 2:4].view(-1,1,1,2).repeat(1,T,V,1)# N*T*V*2
# change to global
z_global = dec_out_all[:, :, :, 2]# N*T*V
z_global[:, :, 0] = root[:, :, 0, 2]
z_global[:, :, 1:] = dec_out_all[:, :, 1:, 2] + root[:, :, 1:, 2] # N*T*V
z_global = z_global.unsqueeze(-1) # N*T*V*1
uv = enc_in_all - cam_c_all # N*T*V*2
xy = uv * z_global.repeat(1, 1, 1, 2) / cam_f_all # N*T*V*2
xyz_global = torch.cat((xy, z_global), -1) # N*T*V*3
xyz_offset = (xyz_global - xyz_global[:, :, 0, :].unsqueeze(-2).repeat(1, 1, V, 1))# N*T*V*3
return xyz_offset |