159 lines
6.7 KiB
Python
159 lines
6.7 KiB
Python
## Our model was revised from https://github.com/zczcwh/PoseFormer/blob/main/common/model_poseformer.py
|
|
|
|
import torch
|
|
import torch.nn as nn
|
|
from functools import partial
|
|
from einops import rearrange
|
|
from timm.models.layers import DropPath
|
|
|
|
from common.opt import opts
|
|
|
|
opt = opts().parse()
|
|
|
|
|
|
#######################################################################################################################
|
|
class Mlp(nn.Module):
|
|
def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.):
|
|
super().__init__()
|
|
out_features = out_features or in_features
|
|
hidden_features = hidden_features or in_features
|
|
self.fc1 = nn.Linear(in_features, hidden_features)
|
|
self.act = act_layer()
|
|
self.fc2 = nn.Linear(hidden_features, out_features)
|
|
self.drop = nn.Dropout(drop)
|
|
|
|
def forward(self, x):
|
|
x = self.fc1(x)
|
|
x = self.act(x)
|
|
x = self.drop(x)
|
|
x = self.fc2(x)
|
|
x = self.drop(x)
|
|
return x
|
|
|
|
|
|
#######################################################################################################################
|
|
class Attention(nn.Module):
|
|
def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0.):
|
|
super().__init__()
|
|
self.num_heads = num_heads
|
|
head_dim = dim // num_heads
|
|
# NOTE scale factor was wrong in my original version, can set manually to be compat with prev weights
|
|
self.scale = qk_scale or head_dim ** -0.5
|
|
|
|
self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
|
|
self.attn_drop = nn.Dropout(attn_drop)
|
|
self.proj = nn.Linear(dim, dim)
|
|
self.proj_drop = nn.Dropout(proj_drop)
|
|
|
|
def forward(self, x):
|
|
B, N, C = x.shape
|
|
qkv = self.qkv(x).reshape(B, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
|
|
q, k, v = qkv[0], qkv[1], qkv[2] # make torchscript happy (cannot use tensor as tuple)
|
|
|
|
attn = (q @ k.transpose(-2, -1)) * self.scale
|
|
attn = attn.softmax(dim=-1)
|
|
attn = self.attn_drop(attn)
|
|
|
|
x = (attn @ v).transpose(1, 2).reshape(B, N, C)
|
|
x = self.proj(x)
|
|
x = self.proj_drop(x)
|
|
return x
|
|
|
|
|
|
#######################################################################################################################
|
|
class CVA_Attention(nn.Module):
|
|
def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0.):
|
|
super().__init__()
|
|
self.num_heads = num_heads
|
|
head_dim = dim // num_heads
|
|
# NOTE scale factor was wrong in my original version, can set manually to be compat with prev weights
|
|
self.scale = qk_scale or head_dim ** -0.5
|
|
|
|
self.Qnorm = nn.LayerNorm(dim)
|
|
self.Knorm = nn.LayerNorm(dim)
|
|
self.Vnorm = nn.LayerNorm(dim)
|
|
self.QLinear = nn.Linear(dim, dim)
|
|
self.KLinear = nn.Linear(dim, dim)
|
|
self.VLinear = nn.Linear(dim, dim)
|
|
|
|
self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
|
|
self.attn_drop = nn.Dropout(attn_drop)
|
|
self.proj = nn.Linear(dim, dim)
|
|
self.proj_drop = nn.Dropout(proj_drop)
|
|
|
|
|
|
|
|
|
|
def forward(self, x, CVA_input):
|
|
B, N, C = x.shape
|
|
q = self.QLinear(self.Qnorm(CVA_input)).reshape(B, N, self.num_heads, C // self.num_heads).permute(0, 2, 1, 3)
|
|
k = self.KLinear(self.Knorm(CVA_input)).reshape(B, N, self.num_heads, C // self.num_heads).permute(0, 2, 1, 3)
|
|
v = self.VLinear(self.Vnorm(x)).reshape(B, N, self.num_heads, C // self.num_heads).permute(0, 2, 1, 3)
|
|
attn = (q @ k.transpose(-2, -1)) * self.scale
|
|
attn = attn.softmax(dim=-1)
|
|
attn = self.attn_drop(attn)
|
|
|
|
x = (attn @ v).transpose(1, 2).reshape(B, N, C)
|
|
x = self.proj(x)
|
|
x = self.proj_drop(x)
|
|
return x
|
|
|
|
|
|
#######################################################################################################################
|
|
class Block(nn.Module):
|
|
|
|
def __init__(self, dim, num_heads, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop=0., attn_drop=0.,
|
|
drop_path=0., act_layer=nn.GELU, norm_layer=nn.LayerNorm):
|
|
super().__init__()
|
|
self.norm1 = norm_layer(dim)
|
|
self.attn = Attention(
|
|
dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=drop)
|
|
# NOTE: drop path for stochastic depth, we shall see if this is better than dropout here
|
|
self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
|
|
self.norm2 = norm_layer(dim)
|
|
mlp_hidden_dim = int(dim * mlp_ratio)
|
|
self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)
|
|
|
|
def forward(self, x):
|
|
x = x + self.drop_path(self.attn(self.norm1(x)))
|
|
x = x + self.drop_path(self.mlp(self.norm2(x)))
|
|
return x
|
|
|
|
|
|
#######################################################################################################################
|
|
class Temporal__features(nn.Module):
|
|
def __init__(self, num_frame=9, num_joints=17, in_chans=2, embed_dim_ratio=32, depth=4,
|
|
num_heads=8, mlp_ratio=2., qkv_bias=True, qk_scale=None,
|
|
drop_rate=0., attn_drop_rate=0., drop_path_rate=0.2, norm_layer=None):
|
|
super().__init__()
|
|
norm_layer = norm_layer or partial(nn.LayerNorm, eps=1e-6)
|
|
embed_dim = embed_dim_ratio * num_joints #### temporal embed_dim is num_joints * spatial embedding dim ratio
|
|
out_dim = num_joints * 3 #### output dimension is num_joints * 3
|
|
### Temporal patch embedding
|
|
self.Temporal_pos_embed = nn.Parameter(torch.zeros(1, num_frame, embed_dim))
|
|
self.pos_drop = nn.Dropout(p=drop_rate)
|
|
|
|
dpr = [x.item() for x in torch.linspace(0, drop_path_rate, depth)] # stochastic depth decay rule
|
|
|
|
self.blocks = nn.ModuleList([
|
|
Block(
|
|
dim=embed_dim, num_heads=num_heads, mlp_ratio=mlp_ratio, qkv_bias=qkv_bias, qk_scale=qk_scale,
|
|
drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[i], norm_layer=norm_layer)
|
|
for i in range(depth)])
|
|
|
|
self.Temporal_norm = norm_layer(embed_dim)
|
|
####### A easy way to implement weighted mean
|
|
self.weighted_mean = torch.nn.Conv1d(in_channels=num_frame, out_channels=1, kernel_size=1)
|
|
|
|
def forward(self, x):
|
|
b = x.shape[0]
|
|
x += self.Temporal_pos_embed
|
|
x = self.pos_drop(x)
|
|
for blk in self.blocks:
|
|
x = blk(x)
|
|
|
|
x = self.Temporal_norm(x)
|
|
##### x size [b, f, emb_dim], then take weighted mean on frame dimension, we only predict 3D pose of the center frame
|
|
# x = self.weighted_mean(x)
|
|
x = x.view(b, opt.frames, -1)
|
|
return x |