|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
""" |
|
|
Various positional encodings for the transformer. |
|
|
""" |
|
|
import math |
|
|
import torch |
|
|
from torch import nn |
|
|
|
|
|
from misc.detr_utils.misc import NestedTensor |
|
|
|
|
|
|
|
|
class PositionEmbeddingSine(nn.Module): |
|
|
""" |
|
|
This is a more standard version of the position embedding, very similar to the one |
|
|
used by the Attention is all you need paper, generalized to work on images. |
|
|
""" |
|
|
def __init__(self, num_pos_feats=64, temperature=10000, normalize=False, scale=None): |
|
|
super().__init__() |
|
|
self.num_pos_feats = num_pos_feats |
|
|
self.temperature = temperature |
|
|
self.normalize = normalize |
|
|
if scale is not None and normalize is False: |
|
|
raise ValueError("normalize should be True if scale is passed") |
|
|
if scale is None: |
|
|
scale = 2 * math.pi |
|
|
self.scale = scale |
|
|
self.max_duration = 256 |
|
|
self.duration_embed_layer = nn.Linear(self.max_duration, self.max_duration) |
|
|
|
|
|
def forward(self, tensor_list: NestedTensor): |
|
|
x = tensor_list.tensors |
|
|
mask = tensor_list.mask |
|
|
duration = tensor_list.duration |
|
|
assert mask is not None |
|
|
not_mask = ~mask |
|
|
x_embed = not_mask.cumsum(1, dtype=torch.float32) |
|
|
if self.normalize: |
|
|
eps = 1e-6 |
|
|
x_embed = (x_embed - 0.5) / (x_embed[:, -1:] + eps) * self.scale |
|
|
|
|
|
dim_t = torch.arange(self.num_pos_feats, dtype=torch.float32, device=x.device) |
|
|
|
|
|
dim_t = self.temperature ** (2 * (torch.div(dim_t, 2, rounding_mode='floor')) / self.num_pos_feats) |
|
|
pos_x = x_embed[:, :, None] / dim_t |
|
|
pos_x = torch.stack((pos_x[:, :, 0::2].sin(), pos_x[:, :, 1::2].cos()), dim=3).flatten(2) |
|
|
|
|
|
dur_embed = self.duration_embedding(duration).reshape(-1,1,self.max_duration).expand_as(pos_x) |
|
|
pos = torch.cat((pos_x, dur_embed), dim=2).permute(0, 2, 1) |
|
|
return pos |
|
|
|
|
|
def duration_embedding(self, durations): |
|
|
out = torch.zeros(len(durations), self.max_duration, device=durations.device) |
|
|
durations = durations.int() |
|
|
for ii in range(len(durations)): |
|
|
out[ii, :durations[ii]] = 1 |
|
|
out = self.duration_embed_layer(out) |
|
|
return out |
|
|
|
|
|
|
|
|
|
|
|
def build_position_encoding(position_embedding, N_steps): |
|
|
if position_embedding in ('v2', 'sine'): |
|
|
|
|
|
position_embedding = PositionEmbeddingSine(N_steps, normalize=True) |
|
|
else: |
|
|
raise ValueError(f"not supported {position_embedding}") |
|
|
|
|
|
return position_embedding |
|
|
|
