forked from cytsinghua/NAUS
-
Notifications
You must be signed in to change notification settings - Fork 0
/
customized_ctc.py
executable file
·155 lines (127 loc) · 8.91 KB
/
customized_ctc.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
# TODO: try to replace fancy tensor indexing by gather / scatter
import math
import torch
import torch.nn.functional as F
# @torch.jit.script
def ctc_loss(log_probs: torch.Tensor, targets: torch.Tensor, input_lengths: torch.Tensor, target_lengths: torch.Tensor,
blank: int = 0, reduction: str = 'none', finfo_min_fp32: float = torch.finfo(torch.float32).min,
finfo_min_fp16: float = torch.finfo(torch.float16).min, alignment: bool = False):
input_time_size, batch_size = log_probs.shape[:2]
B = torch.arange(batch_size, device=input_lengths.device)
_t_a_r_g_e_t_s_ = torch.cat([targets, targets[:, :1]], dim=-1)
_t_a_r_g_e_t_s_ = torch.stack([torch.full_like(_t_a_r_g_e_t_s_, blank), _t_a_r_g_e_t_s_], dim=-1).flatten(
start_dim=-2)
diff_labels = torch.cat([torch.as_tensor([[False, False]], device=targets.device).expand(batch_size, -1),
_t_a_r_g_e_t_s_[:, 2:] != _t_a_r_g_e_t_s_[:, :-2]], dim=1)
# if zero = float('-inf') is used as neutral element, custom logsumexp must be used to avoid nan grad in torch.logsumexp
zero_padding, zero = 2, torch.tensor(finfo_min_fp16 if log_probs.dtype == torch.float16 else finfo_min_fp32,
device=log_probs.device, dtype=log_probs.dtype)
log_probs_ = log_probs.gather(-1, _t_a_r_g_e_t_s_.expand(input_time_size, -1, -1))
log_alpha = torch.full((input_time_size, batch_size, zero_padding + _t_a_r_g_e_t_s_.shape[-1]), zero,
device=log_probs.device, dtype=log_probs.dtype)
log_alpha[0, :, zero_padding + 0] = log_probs[0, :, blank]
log_alpha[0, :, zero_padding + 1] = log_probs[0, B, _t_a_r_g_e_t_s_[:, 1]]
# log_alpha[1:, :, zero_padding:] = log_probs.gather(-1, _t_a_r_g_e_t_s_.expand(len(log_probs), -1, -1))[1:]
is_blank_mask = _t_a_r_g_e_t_s_[0] == blank
is_blank_mask = is_blank_mask.expand(log_alpha[0, :, 2:].shape)
for t in range(1, input_time_size):
log_alpha[t, :, 2:] = log_probs_[t].clone() + logadd(log_alpha[t - 1, :, 2:].clone(), log_alpha[t - 1, :, 1:-1].clone(),
torch.where(diff_labels, log_alpha[t - 1, :, :-2].clone(), zero))
# log_alpha[t, :, 2:] = log_probs_[t].clone() + logadd(torch.where(is_blank_mask, log_alpha[t - 1, :, 2:].clone(), zero),
# log_alpha[t - 1, :, 1:-1].clone(),
# torch.where(diff_labels, log_alpha[t - 1, :, :-2].clone(), zero))
l1l2 = log_alpha[input_lengths - 1, B].gather(-1, torch.stack(
[zero_padding + target_lengths * 2 - 1, zero_padding + target_lengths * 2], dim=-1))
loss = -torch.logsumexp(l1l2, dim=-1)
return loss
if not alignment:
return loss
# below is for debugging, for real alignment use more efficient the distinct ctc_alignment(...) method
path = torch.zeros(len(log_alpha), len(B), device=log_alpha.device, dtype=torch.int64)
path[input_lengths - 1, B] = zero_padding + 2 * target_lengths - 1 + l1l2.max(dim=-1).indices
for t, indices in reversed(list(enumerate(path))[1:]):
indices_ = torch.stack(
[(indices - 2) * diff_labels[B, (indices - zero_padding).clamp(min=0)], (indices - 1).clamp(min=0),
indices], dim=-1)
path[t - 1] += (indices - 2 + log_alpha[t - 1, B].gather(-1, indices_).max(dim=-1).indices).clamp(min=0)
return torch.zeros_like(log_alpha).scatter_(-1, path.unsqueeze(-1), 1.0)[..., (zero_padding + 1)::2]
@torch.jit.script
def ctc_alignment(log_probs: torch.Tensor, targets: torch.Tensor, input_lengths: torch.Tensor,
target_lengths: torch.Tensor, blank: int = 0, finfo_min_fp32: float = torch.finfo(torch.float32).min,
finfo_min_fp16: float = torch.finfo(torch.float16).min):
input_time_size, batch_size = log_probs.shape[:2]
B = torch.arange(batch_size, device=input_lengths.device)
_t_a_r_g_e_t_s_ = torch.cat([
torch.stack([torch.full_like(targets, blank), targets], dim=-1).flatten(start_dim=-2),
torch.full_like(targets[:, :1], blank)
], dim=-1)
diff_labels = torch.cat([
torch.as_tensor([[False, False]], device=targets.device).expand(batch_size, -1),
_t_a_r_g_e_t_s_[:, 2:] != _t_a_r_g_e_t_s_[:, :-2]
], dim=1)
zero_padding, zero = 2, torch.tensor(finfo_min_fp16 if log_probs.dtype == torch.float16 else finfo_min_fp32,
device=log_probs.device, dtype=log_probs.dtype)
padded_t = zero_padding + _t_a_r_g_e_t_s_.shape[-1]
log_alpha = torch.full((batch_size, padded_t), zero, device=log_probs.device, dtype=log_probs.dtype)
log_alpha[:, zero_padding + 0] = log_probs[0, :, blank]
log_alpha[:, zero_padding + 1] = log_probs[0, B, _t_a_r_g_e_t_s_[:, 1]]
packmask = 0b11
packnibbles = 4 # packnibbles = 1
backpointers_shape = [len(log_probs), batch_size, int(math.ceil(padded_t / packnibbles))]
backpointers = torch.zeros(backpointers_shape, device=log_probs.device, dtype=torch.uint8)
backpointer = torch.zeros(backpointers_shape[1:], device=log_probs.device, dtype=torch.uint8)
packshift = torch.tensor([[[6, 4, 2, 0]]], device=log_probs.device, dtype=torch.uint8)
for t in range(1, input_time_size):
prev = torch.stack([log_alpha[:, 2:], log_alpha[:, 1:-1], torch.where(diff_labels, log_alpha[:, :-2], zero)])
log_alpha[:, zero_padding:] = log_probs[t].gather(-1, _t_a_r_g_e_t_s_) + prev.logsumexp(dim=0)
backpointer[:, zero_padding:(zero_padding + prev.shape[-1])] = prev.argmax(dim=0)
torch.sum(backpointer.view(len(backpointer), -1, packnibbles) << packshift, dim=-1,
out=backpointers[t]) # backpointers[t] = backpointer
l1l2 = log_alpha.gather(-1, torch.stack([zero_padding + target_lengths * 2 - 1, zero_padding + target_lengths * 2],
dim=-1))
path = torch.zeros(input_time_size, batch_size, device=log_alpha.device, dtype=torch.long)
path[input_lengths - 1, B] = zero_padding + target_lengths * 2 - 1 + l1l2.argmax(dim=-1)
for t in range(input_time_size - 1, 0, -1):
indices = path[t]
backpointer = (backpointers[t].unsqueeze(-1) >> packshift).view_as(backpointer) # backpointer = backpointers[t]
path[t - 1] += indices - backpointer.gather(-1, indices.unsqueeze(-1)).squeeze(-1).bitwise_and_(packmask)
return torch.zeros_like(_t_a_r_g_e_t_s_, dtype=torch.int64).scatter_(-1, (path.t() - zero_padding).clamp(min=0),
torch.arange(input_time_size,
device=log_alpha.device).expand(
batch_size, -1))[:, 1::2]
def ctc_alignment_targets(log_probs, targets, input_lengths, target_lengths, blank=0, ctc_loss=F.ctc_loss,
retain_graph=True):
loss = ctc_loss(log_probs, targets, input_lengths, target_lengths, blank=blank, reduction='sum')
probs = log_probs.exp()
# to simplify API we inline log_softmax gradient, i.e. next two lines are equivalent to: grad_logits, = torch.autograd.grad(loss, logits, retain_graph = True). gradient formula explained at https://stackoverflow.com/questions/35304393/trying-to-understand-code-that-computes-the-gradient-wrt-to-the-input-for-logsof
grad_log_probs, = torch.autograd.grad(loss, log_probs, retain_graph=retain_graph)
grad_logits = grad_log_probs - probs * grad_log_probs.sum(dim=-1, keepdim=True)
temporal_mask = (torch.arange(len(log_probs), device=input_lengths.device, dtype=input_lengths.dtype).unsqueeze(
1) < input_lengths.unsqueeze(0)).unsqueeze(-1)
return (probs * temporal_mask - grad_logits).detach()
def logadd(x0, x1, x2):
# produces nan gradients in backward if -inf log-space zero element is used https://github.com/pytorch/pytorch/issues/31829
return torch.logsumexp(torch.stack([x0, x1, x2]), dim=0)
# use if -inf log-space zero element is used
# return LogsumexpFunction.apply(x0, x1, x2)
# produces inplace modification error https://github.com/pytorch/pytorch/issues/31819
# m = torch.max(torch.max(x0, x1), x2)
# m = m.masked_fill(torch.isinf(m), 0)
# res = (x0 - m).exp() + (x1 - m).exp() + (x2 - m).exp()
# return res.log().add(m)
class LogsumexpFunction(torch.autograd.function.Function):
@staticmethod
def forward(self, x0, x1, x2):
m = torch.max(torch.max(x0, x1), x2)
m = m.masked_fill_(torch.isinf(m), 0)
e0 = (x0 - m).exp_()
e1 = (x1 - m).exp_()
e2 = (x2 - m).exp_()
e = (e0 + e1).add_(e2).clamp_(min=1e-16)
self.save_for_backward(e0, e1, e2, e)
return e.log_().add_(m)
@staticmethod
def backward(self, grad_output):
e0, e1, e2, e = self.saved_tensors
g = grad_output / e
return g * e0, g * e1, g * e2