/
loss_functions.py
155 lines (123 loc) · 5.75 KB
/
loss_functions.py
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
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
from __future__ import division
import numpy as np
import torch
from torch import nn
import torch.nn.functional as F
from inverse_warp import inverse_warp
def photometric_reconstruction_loss(tgt_img, ref_imgs, intrinsics,
depth, explainability_mask, pose,
rotation_mode='euler', padding_mode='zeros'):
def one_scale(depth, explainability_mask):
assert(explainability_mask is None or depth.size()[2:] == explainability_mask.size()[2:])
assert(pose.size(1) == len(ref_imgs))
reconstruction_loss = 0
b, _, h, w = depth.size()
downscale = tgt_img.size(2)/h
tgt_img_scaled = F.interpolate(tgt_img, (h, w), mode='area')
ref_imgs_scaled = [F.interpolate(ref_img, (h, w), mode='area') for ref_img in ref_imgs]
intrinsics_scaled = torch.cat((intrinsics[:, 0:2]/downscale, intrinsics[:, 2:]), dim=1)
warped_imgs = []
diff_maps = []
for i, ref_img in enumerate(ref_imgs_scaled):
current_pose = pose[:, i]
ref_img_warped, valid_points = inverse_warp(ref_img, depth[:,0], current_pose,
intrinsics_scaled,
rotation_mode, padding_mode)
diff = (tgt_img_scaled - ref_img_warped) * valid_points.unsqueeze(1).float()
if explainability_mask is not None:
diff = diff * explainability_mask[:,i:i+1].expand_as(diff)
reconstruction_loss += diff.abs().mean()
assert((reconstruction_loss == reconstruction_loss).item() == 1)
warped_imgs.append(ref_img_warped[0])
diff_maps.append(diff[0])
return reconstruction_loss, warped_imgs, diff_maps
warped_results, diff_results = [], []
if type(explainability_mask) not in [tuple, list]:
explainability_mask = [explainability_mask]
if type(depth) not in [list, tuple]:
depth = [depth]
total_loss = 0
for d, mask in zip(depth, explainability_mask):
loss, warped, diff = one_scale(d, mask)
total_loss += loss
warped_results.append(warped)
diff_results.append(diff)
return total_loss, warped_results, diff_results
def explainability_loss(mask):
if type(mask) not in [tuple, list]:
mask = [mask]
loss = 0
for mask_scaled in mask:
ones_var = torch.ones_like(mask_scaled)
loss += nn.functional.binary_cross_entropy(mask_scaled, ones_var)
return loss
def smooth_loss(pred_map):
def gradient(pred):
D_dy = pred[:, :, 1:] - pred[:, :, :-1]
D_dx = pred[:, :, :, 1:] - pred[:, :, :, :-1]
return D_dx, D_dy
if type(pred_map) not in [tuple, list]:
pred_map = [pred_map]
loss = 0
weight = 1.
for scaled_map in pred_map:
dx, dy = gradient(scaled_map)
dx2, dxdy = gradient(dx)
dydx, dy2 = gradient(dy)
loss += (dx2.abs().mean() + dxdy.abs().mean() + dydx.abs().mean() + dy2.abs().mean())*weight
weight /= 2.3 # don't ask me why it works better
return loss
@torch.no_grad()
def compute_depth_errors(gt, pred, crop=True):
abs_diff, abs_rel, sq_rel, a1, a2, a3 = 0,0,0,0,0,0
batch_size = gt.size(0)
'''
crop used by Garg ECCV16 to reprocude Eigen NIPS14 results
construct a mask of False values, with the same size as target
and then set to True values inside the crop
'''
if crop:
crop_mask = gt[0] != gt[0]
y1,y2 = int(0.40810811 * gt.size(1)), int(0.99189189 * gt.size(1))
x1,x2 = int(0.03594771 * gt.size(2)), int(0.96405229 * gt.size(2))
crop_mask[y1:y2,x1:x2] = 1
skipped = 0
for current_gt, current_pred in zip(gt, pred):
valid = (current_gt > 0) & (current_gt < 80)
if crop:
valid = valid & crop_mask
if valid.sum() == 0:
continue
valid_gt = current_gt[valid]
valid_pred = current_pred[valid]
median_pred = torch.median(valid_pred.clamp(1e-3, 80))
valid_pred = valid_pred * torch.median(valid_gt) / median_pred
valid_pred = valid_pred.clamp(1e-3, 80)
thresh = torch.max((valid_gt / valid_pred), (valid_pred / valid_gt))
a1 += (thresh < 1.25).float().mean()
a2 += (thresh < 1.25 ** 2).float().mean()
a3 += (thresh < 1.25 ** 3).float().mean()
abs_diff += torch.mean(torch.abs(valid_gt - valid_pred))
abs_rel += torch.mean(torch.abs(valid_gt - valid_pred) / valid_gt)
sq_rel += torch.mean(((valid_gt - valid_pred)**2) / valid_gt)
if skipped == batch_size:
return None
return [metric.item() / (batch_size - skipped) for metric in [abs_diff, abs_rel, sq_rel, a1, a2, a3]]
@torch.no_grad()
def compute_pose_errors(gt, pred):
RE = 0
for (current_gt, current_pred) in zip(gt, pred):
snippet_length = current_gt.shape[0]
scale_factor = torch.sum(current_gt[..., -1] * current_pred[..., -1]) / torch.sum(current_pred[..., -1] ** 2)
ATE = torch.norm((current_gt[..., -1] - scale_factor * current_pred[..., -1]).reshape(-1)).cpu().numpy()
R = current_gt[..., :3] @ current_pred[..., :3].transpose(-2, -1)
for gt_pose, pred_pose in zip(current_gt, current_pred):
# Residual matrix to which we compute angle's sin and cos
R = (gt_pose[:, :3] @ torch.inverse(pred_pose[:, :3])).cpu().numpy()
s = np.linalg.norm([R[0, 1]-R[1, 0],
R[1, 2]-R[2, 1],
R[0, 2]-R[2, 0]])
c = np.trace(R) - 1
# Note: we actually compute double of cos and sin, but arctan2 is invariant to scale
RE += np.arctan2(s, c)
return [ATE/snippet_length, RE/snippet_length]