You can not select more than 25 topics
Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
494 lines
18 KiB
494 lines
18 KiB
# Copyright (c) OpenMMLab. All rights reserved.
|
|
import math
|
|
|
|
import torch
|
|
from functools import partial
|
|
import torch.nn as nn
|
|
import torch.nn.functional as F
|
|
import torch.utils.checkpoint as checkpoint
|
|
|
|
from timm.models.layers import drop_path, to_2tuple, trunc_normal_
|
|
|
|
from ..builder import BACKBONES
|
|
from .base_backbone import BaseBackbone
|
|
|
|
from .sam_vit import build_vit_sam
|
|
|
|
|
|
def get_abs_pos(abs_pos, h, w, ori_h, ori_w, has_cls_token=True):
|
|
"""
|
|
Calculate absolute positional embeddings. If needed, resize embeddings and remove cls_token
|
|
dimension for the original embeddings.
|
|
Args:
|
|
abs_pos (Tensor): absolute positional embeddings with (1, num_position, C).
|
|
has_cls_token (bool): If true, has 1 embedding in abs_pos for cls token.
|
|
hw (Tuple): size of input image tokens.
|
|
|
|
Returns:
|
|
Absolute positional embeddings after processing with shape (1, H, W, C)
|
|
"""
|
|
cls_token = None
|
|
B, L, C = abs_pos.shape
|
|
if has_cls_token:
|
|
cls_token = abs_pos[:, 0:1]
|
|
abs_pos = abs_pos[:, 1:]
|
|
|
|
if ori_h != h or ori_w != w:
|
|
new_abs_pos = F.interpolate(
|
|
abs_pos.reshape(1, ori_h, ori_w, -1).permute(0, 3, 1, 2),
|
|
size=(h, w),
|
|
mode="bicubic",
|
|
align_corners=False,
|
|
).permute(0, 2, 3, 1).reshape(B, -1, C)
|
|
|
|
else:
|
|
new_abs_pos = abs_pos
|
|
|
|
if cls_token is not None:
|
|
new_abs_pos = torch.cat([cls_token, new_abs_pos], dim=1)
|
|
return new_abs_pos
|
|
|
|
class DropPath(nn.Module):
|
|
"""Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks).
|
|
"""
|
|
def __init__(self, drop_prob=None):
|
|
super(DropPath, self).__init__()
|
|
self.drop_prob = drop_prob
|
|
|
|
def forward(self, x):
|
|
return drop_path(x, self.drop_prob, self.training)
|
|
|
|
def extra_repr(self):
|
|
return 'p={}'.format(self.drop_prob)
|
|
|
|
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.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., attn_head_dim=None,):
|
|
super().__init__()
|
|
self.num_heads = num_heads
|
|
head_dim = dim // num_heads
|
|
self.dim = dim
|
|
|
|
if attn_head_dim is not None:
|
|
head_dim = attn_head_dim
|
|
all_head_dim = head_dim * self.num_heads
|
|
|
|
self.scale = qk_scale or head_dim ** -0.5
|
|
|
|
self.qkv = nn.Linear(dim, all_head_dim * 3, bias=qkv_bias)
|
|
|
|
self.attn_drop = nn.Dropout(attn_drop)
|
|
self.proj = nn.Linear(all_head_dim, dim)
|
|
self.proj_drop = nn.Dropout(proj_drop)
|
|
|
|
def forward(self, x):
|
|
B, N, C = x.shape
|
|
qkv = self.qkv(x)
|
|
qkv = qkv.reshape(B, N, 3, self.num_heads, -1).permute(2, 0, 3, 1, 4)
|
|
q, k, v = qkv[0], qkv[1], qkv[2] # make torchscript happy (cannot use tensor as tuple)
|
|
|
|
q = q * self.scale
|
|
attn = (q @ k.transpose(-2, -1))
|
|
|
|
attn = attn.softmax(dim=-1)
|
|
attn = self.attn_drop(attn)
|
|
|
|
x = (attn @ v).transpose(1, 2).reshape(B, N, -1)
|
|
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, attn_head_dim=None
|
|
):
|
|
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, attn_head_dim=attn_head_dim
|
|
)
|
|
|
|
# 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 PatchEmbed(nn.Module):
|
|
""" Image to Patch Embedding
|
|
"""
|
|
def __init__(self, img_size=224, patch_size=16, in_chans=3, embed_dim=768, ratio=1):
|
|
super().__init__()
|
|
img_size = to_2tuple(img_size)
|
|
patch_size = to_2tuple(patch_size)
|
|
num_patches = (img_size[1] // patch_size[1]) * (img_size[0] // patch_size[0]) * (ratio ** 2)
|
|
self.patch_shape = (int(img_size[0] // patch_size[0] * ratio), int(img_size[1] // patch_size[1] * ratio))
|
|
self.origin_patch_shape = (int(img_size[0] // patch_size[0]), int(img_size[1] // patch_size[1]))
|
|
self.img_size = img_size
|
|
self.patch_size = patch_size
|
|
self.num_patches = num_patches
|
|
|
|
self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=(patch_size[0] // ratio), padding=4 + 2 * (ratio//2-1))
|
|
|
|
def forward(self, x, **kwargs):
|
|
B, C, H, W = x.shape
|
|
x = self.proj(x)
|
|
Hp, Wp = x.shape[2], x.shape[3]
|
|
|
|
x = x.flatten(2).transpose(1, 2)
|
|
return x, (Hp, Wp)
|
|
|
|
|
|
class HybridEmbed(nn.Module):
|
|
""" CNN Feature Map Embedding
|
|
Extract feature map from CNN, flatten, project to embedding dim.
|
|
"""
|
|
def __init__(self, backbone, img_size=224, feature_size=None, in_chans=3, embed_dim=768):
|
|
super().__init__()
|
|
assert isinstance(backbone, nn.Module)
|
|
img_size = to_2tuple(img_size)
|
|
self.img_size = img_size
|
|
self.backbone = backbone
|
|
if feature_size is None:
|
|
with torch.no_grad():
|
|
training = backbone.training
|
|
if training:
|
|
backbone.eval()
|
|
o = self.backbone(torch.zeros(1, in_chans, img_size[0], img_size[1]))[-1]
|
|
feature_size = o.shape[-2:]
|
|
feature_dim = o.shape[1]
|
|
backbone.train(training)
|
|
else:
|
|
feature_size = to_2tuple(feature_size)
|
|
feature_dim = self.backbone.feature_info.channels()[-1]
|
|
self.num_patches = feature_size[0] * feature_size[1]
|
|
self.proj = nn.Linear(feature_dim, embed_dim)
|
|
|
|
def forward(self, x):
|
|
x = self.backbone(x)[-1]
|
|
x = x.flatten(2).transpose(1, 2)
|
|
x = self.proj(x)
|
|
return x
|
|
|
|
class Cross_Attention(nn.Module):
|
|
def __init__(self, dim, num_heads=12, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0.):
|
|
super().__init__()
|
|
self.num_heads = num_heads
|
|
head_dim = dim // num_heads
|
|
self.scale = qk_scale or head_dim ** -0.5
|
|
|
|
self.linear_q = nn.Linear(dim, dim, bias=qkv_bias)
|
|
self.linear_k = nn.Linear(dim, dim, bias=qkv_bias)
|
|
self.linear_v = nn.Linear(dim, dim, 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_1, x_2, x_3):
|
|
B, N, C = x_1.shape # q
|
|
B, N_1, C = x_2.shape # k, v
|
|
|
|
q = self.linear_q(x_1).reshape(B, N, self.num_heads, C // self.num_heads).permute(0, 2, 1, 3) # (B, num_heads, N, C//num_heads)
|
|
k = self.linear_k(x_2).reshape(B, N_1, self.num_heads, C // self.num_heads).permute(0, 2, 1, 3) # (B, num_heads, N_1, C//num_heads)
|
|
v = self.linear_v(x_3).reshape(B, N_1, self.num_heads, C // self.num_heads).permute(0, 2, 1, 3) # (B, num_heads, N_1, C//num_heads)
|
|
|
|
attn = (q @ k.transpose(-2, -1)) * self.scale # (B, num_heads, N, N_1)
|
|
attn = attn.softmax(dim=-1)
|
|
attn = self.attn_drop(attn)
|
|
|
|
# import matplotlib.pyplot as plt
|
|
# import seaborn as sns
|
|
|
|
# attn_map = attn[0][0].cpu().detach().numpy()
|
|
# plt.figure(figsize=(20, 10))
|
|
# sns.heatmap(attn_map, annot=True, fmt='.2f', cmap='coolwarm')
|
|
|
|
# plt.title('Cross Attention Map')
|
|
# plt.xlabel('N_1')
|
|
# plt.ylabel('N')
|
|
|
|
# plt.savefig('/home/fhw/code/ViTPose/test/cross_attn_map.png')
|
|
# plt.close()
|
|
|
|
x = (attn @ v).transpose(1, 2).reshape(B, N, C) # (B, N, C)
|
|
x = self.proj(x)
|
|
x = self.proj_drop(x)
|
|
return x
|
|
|
|
class CustomAttentionFFN(nn.Module):
|
|
def __init__(self, dim, num_heads=12, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0.):
|
|
super().__init__()
|
|
self.self_attn = Attention(
|
|
dim, num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale,
|
|
attn_drop=attn_drop, proj_drop=proj_drop)
|
|
|
|
self.cross_attn = Cross_Attention(dim, num_heads=num_heads, qkv_bias=qkv_bias, \
|
|
qk_scale=qk_scale, attn_drop=attn_drop, proj_drop=proj_drop)
|
|
|
|
self.ffn = nn.Sequential(
|
|
nn.Linear(dim, dim * 4),
|
|
nn.GELU(),
|
|
nn.Linear(dim * 4, dim),
|
|
nn.Dropout(proj_drop)
|
|
)
|
|
self.norm1 = nn.LayerNorm(dim)
|
|
self.norm2 = nn.LayerNorm(dim)
|
|
self.norm3 = nn.LayerNorm(dim)
|
|
|
|
def forward(self, x1, x2):
|
|
x1 = self.norm1(x1 + self.self_attn(x1))
|
|
|
|
x1 = self.norm2(x1 + self.cross_attn(x1, x2, x2))
|
|
|
|
x1 = self.norm3(x1 + self.ffn(x1))
|
|
|
|
return x1
|
|
|
|
@BACKBONES.register_module()
|
|
class ViTSam(BaseBackbone):
|
|
|
|
def __init__(self,
|
|
img_size=224, patch_size=16, in_chans=3, num_classes=80, embed_dim=768, depth=12,
|
|
num_heads=12, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop_rate=0., attn_drop_rate=0.,
|
|
drop_path_rate=0., hybrid_backbone=None, norm_layer=None, use_checkpoint=False,
|
|
frozen_stages=-1, ratio=1, last_norm=True,
|
|
patch_padding='pad', freeze_attn=False, freeze_ffn=False,
|
|
samvit_checkpoint=None, sam_img_size=1024
|
|
):
|
|
# Protect mutable default arguments
|
|
super(ViTSam, self).__init__()
|
|
norm_layer = norm_layer or partial(nn.LayerNorm, eps=1e-6)
|
|
self.num_classes = num_classes
|
|
self.num_features = self.embed_dim = embed_dim # num_features for consistency with other models
|
|
self.frozen_stages = frozen_stages
|
|
self.use_checkpoint = use_checkpoint
|
|
self.patch_padding = patch_padding
|
|
self.freeze_attn = freeze_attn
|
|
self.freeze_ffn = freeze_ffn
|
|
self.depth = depth
|
|
|
|
if hybrid_backbone is not None:
|
|
self.patch_embed = HybridEmbed(
|
|
hybrid_backbone, img_size=img_size, in_chans=in_chans, embed_dim=embed_dim)
|
|
else:
|
|
self.patch_embed = PatchEmbed(
|
|
img_size=img_size, patch_size=patch_size, in_chans=in_chans, embed_dim=embed_dim, ratio=ratio)
|
|
num_patches = self.patch_embed.num_patches
|
|
|
|
# since the pretraining model has class token
|
|
self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, embed_dim))
|
|
|
|
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.last_norm = norm_layer(embed_dim) if last_norm else nn.Identity()
|
|
|
|
if self.pos_embed is not None:
|
|
trunc_normal_(self.pos_embed, std=.02)
|
|
|
|
self._freeze_stages()
|
|
|
|
# ======================== SAM-ViT ========================
|
|
self.sam_vit = build_vit_sam(model_name='vit_b', checkpoint=samvit_checkpoint, img_size=sam_img_size)
|
|
self.sam_vit.eval()
|
|
for param in self.sam_vit.parameters():
|
|
param.requires_grad = False
|
|
|
|
# 交叉注意力
|
|
self.cross_attn = Cross_Attention(embed_dim, num_heads=num_heads, qkv_bias=qkv_bias, \
|
|
qk_scale=qk_scale, attn_drop=attn_drop_rate, proj_drop=drop_rate)
|
|
|
|
# vit_token做自注意力后，再和sam_token做交叉注意力，得到的结果再经过FFN
|
|
# self.custom_attn_ffn = CustomAttentionFFN(embed_dim, num_heads=num_heads, qkv_bias=qkv_bias, \
|
|
# qk_scale=qk_scale, attn_drop=attn_drop_rate, proj_drop=drop_rate)
|
|
|
|
# 在sam_encoder后面加一层ffn
|
|
self.sam_ffn = nn.Sequential(
|
|
nn.Linear(embed_dim, embed_dim * 4),
|
|
nn.GELU(),
|
|
nn.Linear(embed_dim * 4, embed_dim),
|
|
nn.Dropout(drop_rate)
|
|
)
|
|
self.sam_norm = norm_layer(embed_dim)
|
|
|
|
def _freeze_stages(self):
|
|
"""Freeze parameters."""
|
|
if self.frozen_stages >= 0:
|
|
self.patch_embed.eval()
|
|
for param in self.patch_embed.parameters():
|
|
param.requires_grad = False
|
|
|
|
for i in range(0, self.frozen_stages):
|
|
m = self.blocks[i]
|
|
m.eval()
|
|
for param in m.parameters():
|
|
param.requires_grad = False
|
|
|
|
if self.freeze_attn:
|
|
for i in range(0, self.depth):
|
|
m = self.blocks[i]
|
|
m.attn.eval()
|
|
m.norm1.eval()
|
|
for param in m.attn.parameters():
|
|
param.requires_grad = False
|
|
for param in m.norm1.parameters():
|
|
param.requires_grad = False
|
|
|
|
if self.freeze_ffn:
|
|
self.pos_embed.requires_grad = False
|
|
self.patch_embed.eval()
|
|
for param in self.patch_embed.parameters():
|
|
param.requires_grad = False
|
|
for i in range(0, self.depth):
|
|
m = self.blocks[i]
|
|
m.mlp.eval()
|
|
m.norm2.eval()
|
|
for param in m.mlp.parameters():
|
|
param.requires_grad = False
|
|
for param in m.norm2.parameters():
|
|
param.requires_grad = False
|
|
|
|
def init_weights(self, pretrained=None):
|
|
"""Initialize the weights in backbone.
|
|
Args:
|
|
pretrained (str, optional): Path to pre-trained weights.
|
|
Defaults to None.
|
|
"""
|
|
super().init_weights(pretrained, patch_padding=self.patch_padding)
|
|
|
|
if pretrained is None:
|
|
def _init_weights(m):
|
|
if isinstance(m, nn.Linear):
|
|
trunc_normal_(m.weight, std=.02)
|
|
if isinstance(m, nn.Linear) and m.bias is not None:
|
|
nn.init.constant_(m.bias, 0)
|
|
elif isinstance(m, nn.LayerNorm):
|
|
nn.init.constant_(m.bias, 0)
|
|
nn.init.constant_(m.weight, 1.0)
|
|
|
|
self.apply(_init_weights)
|
|
|
|
def get_num_layers(self):
|
|
return len(self.blocks)
|
|
|
|
@torch.jit.ignore
|
|
def no_weight_decay(self):
|
|
return {'pos_embed', 'cls_token'}
|
|
|
|
def forward_features(self, x):
|
|
B, C, H, W = x.shape
|
|
x, (Hp, Wp) = self.patch_embed(x)
|
|
|
|
if self.pos_embed is not None:
|
|
# fit for multiple GPU training
|
|
# since the first element for pos embed (sin-cos manner) is zero, it will cause no difference
|
|
x = x + self.pos_embed[:, 1:] + self.pos_embed[:, :1]
|
|
|
|
for blk in self.blocks:
|
|
if self.use_checkpoint:
|
|
x = checkpoint.checkpoint(blk, x)
|
|
else:
|
|
x = blk(x)
|
|
|
|
x = self.last_norm(x)
|
|
|
|
return x, Hp, Wp
|
|
|
|
def forward(self, x1, x2):
|
|
import time
|
|
B, _, _, _ = x1.shape
|
|
x1, Hp, Wp = self.forward_features(x1) # B, N_vitpose, C
|
|
|
|
with torch.no_grad():
|
|
# start_time = time.time()
|
|
# self.sam_vit.eval()
|
|
x2 = self.sam_vit(x2) # B, N_sam, C
|
|
|
|
# end_time = time.time()
|
|
# print('SAM-ViT forward time: {:.4f}秒'.format(end_time - start_time))
|
|
|
|
x2 = self.sam_norm(x2 + self.sam_ffn(x2))
|
|
|
|
x1 = x1 + self.cross_attn(x1, x2, x2)
|
|
|
|
# x1 = self.custom_attn_ffn(x1, x2)
|
|
|
|
xp = x1.permute(0, 2, 1).reshape(B, -1, Hp, Wp).contiguous() # B, C, Hp, Wp
|
|
return xp
|
|
|
|
def train(self, mode=True):
|
|
"""Convert the model into training mode."""
|
|
super().train(mode)
|
|
self._freeze_stages()
|
|
|
|
|
|
if __name__ == '__main__':
|
|
from thop import profile
|
|
from mmcv.runner import load_checkpoint
|
|
|
|
# Prepare an image as the input
|
|
bs, c, h, w = 2, 3, 1024, 1024
|
|
x1 = torch.randn(bs, c, 256, 192)
|
|
x2 = torch.randn(bs, c, h, w)
|
|
patch_size = 16
|
|
device = torch.device('cuda:1' if torch.cuda.is_available() else 'cpu')
|
|
|
|
# Build model
|
|
model = ViTSam(img_size=(256, 192), patch_size=16, embed_dim=768, depth=12, num_heads=12, ratio=1,
|
|
use_checkpoint=False, mlp_ratio=4, qkv_bias=True, drop_path_rate=0.3,
|
|
samvit_checkpoint='/home/fhw/code/ViTPose/checkpoints/sam/sam_vit_b_01ec64.pth')
|
|
|
|
|
|
if torch.cuda.is_available():
|
|
x1 = x1.to(device)
|
|
x2 = x2.to(device)
|
|
model = model.to(device)
|
|
|
|
with torch.no_grad():
|
|
model.eval()
|
|
# Inference
|
|
outputs = model(x1, x2)
|
|
print(outputs.shape)
|
|
|
|
# Compute FLOPs & Params
|
|
print('==============================')
|
|
model.eval()
|
|
flops, params = profile(model, inputs=(x1, x2), verbose=False)
|
|
print('GFLOPs : {:.2f}'.format(flops / 1e9 * 2))
|
|
print('Params : {:.2f} M'.format(params / 1e6))
|
|
|