Spaces:
Sleeping
Sleeping
File size: 1,283 Bytes
8e0b903 |
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 |
import math
from dataclasses import dataclass
import torch
import torch.nn as nn
from einops import rearrange, repeat
from x3D_utils import BaseModule
class Triplane1DTokenizer(BaseModule):
@dataclass
class Config(BaseModule.Config):
plane_size: int
num_channels: int
cfg: Config
def configure(self) -> None:
self.embeddings = nn.Parameter(
torch.randn(
(3, self.cfg.num_channels, self.cfg.plane_size, self.cfg.plane_size),
dtype=torch.float32,
)
* 1
/ math.sqrt(self.cfg.num_channels)
)
def forward(self, batch_size: int) -> torch.Tensor:
return rearrange(
repeat(self.embeddings, "Np Ct Hp Wp -> B Np Ct Hp Wp", B=batch_size),
"B Np Ct Hp Wp -> B Ct (Np Hp Wp)",
)
def detokenize(self, tokens: torch.Tensor) -> torch.Tensor:
batch_size, Ct, Nt = tokens.shape
assert Nt == self.cfg.plane_size**2 * 3
assert Ct == self.cfg.num_channels
return rearrange(
tokens,
"B Ct (Np Hp Wp) -> B Np Ct Hp Wp",
Np=3,
Hp=self.cfg.plane_size,
Wp=self.cfg.plane_size,
)
|