"""
Modified from https://github.com/Sense-X/UniFormer/blob/main/image_classification/models/uniformer.py
"""
from functools import partial
from collections import OrderedDict
import math
import oneflow as flow
import oneflow.nn as nn
import oneflow.nn.functional as F
from flowvision.data import IMAGENET_INCEPTION_MEAN, IMAGENET_INCEPTION_STD
from flowvision.layers import trunc_normal_, DropPath
from .helpers import to_2tuple
from .registry import ModelCreator
from .utils import load_state_dict_from_url
model_urls = {
"uniformer_base": "https://oneflow-public.oss-cn-beijing.aliyuncs.com/model_zoo/flowvision/classification/UniFormer/uniformer_base_oneflow.zip",
"uniformer_base_ls": "https://oneflow-public.oss-cn-beijing.aliyuncs.com/model_zoo/flowvision/classification/UniFormer/uniformer_base_ls_oneflow.zip",
"uniformer_small": "https://oneflow-public.oss-cn-beijing.aliyuncs.com/model_zoo/flowvision/classification/UniFormer/uniformer_small_oneflow.zip",
"uniformer_small_plus": "https://oneflow-public.oss-cn-beijing.aliyuncs.com/model_zoo/flowvision/classification/UniFormer/uniformer_small_plus_oneflow.zip",
}
layer_scale = False
init_value = 1e-6
class Mlp(nn.Module):
def __init__(
self,
in_features,
hidden_features=None,
out_features=None,
act_layer=nn.GELU,
drop=0.0,
):
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 CMlp(nn.Module):
def __init__(
self,
in_features,
hidden_features=None,
out_features=None,
act_layer=nn.GELU,
drop=0.0,
):
super().__init__()
out_features = out_features or in_features
hidden_features = hidden_features or in_features
self.fc1 = nn.Conv2d(in_features, hidden_features, 1)
self.act = act_layer()
self.fc2 = nn.Conv2d(hidden_features, out_features, 1)
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=0.0,
proj_drop=0.0,
):
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]
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 CBlock(nn.Module):
def __init__(
self,
dim,
num_heads,
mlp_ratio=4.0,
qkv_bias=False,
qk_scale=None,
drop=0.0,
attn_drop=0.0,
drop_path=0.0,
act_layer=nn.GELU,
norm_layer=nn.LayerNorm,
):
super().__init__()
self.pos_embed = nn.Conv2d(dim, dim, 3, padding=1, groups=dim)
self.norm1 = nn.BatchNorm2d(dim)
self.conv1 = nn.Conv2d(dim, dim, 1)
self.conv2 = nn.Conv2d(dim, dim, 1)
self.attn = nn.Conv2d(dim, dim, 5, padding=2, groups=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 > 0.0 else nn.Identity()
self.norm2 = nn.BatchNorm2d(dim)
mlp_hidden_dim = int(dim * mlp_ratio)
self.mlp = CMlp(
in_features=dim,
hidden_features=mlp_hidden_dim,
act_layer=act_layer,
drop=drop,
)
def forward(self, x):
x = x + self.pos_embed(x)
x = x + self.drop_path(self.conv2(self.attn(self.conv1(self.norm1(x)))))
x = x + self.drop_path(self.mlp(self.norm2(x)))
return x
class SABlock(nn.Module):
def __init__(
self,
dim,
num_heads,
mlp_ratio=4.0,
qkv_bias=False,
qk_scale=None,
drop=0.0,
attn_drop=0.0,
drop_path=0.0,
act_layer=nn.GELU,
norm_layer=nn.LayerNorm,
):
super().__init__()
self.pos_embed = nn.Conv2d(dim, dim, 3, padding=1, groups=dim)
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 > 0.0 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,
)
global layer_scale
self.ls = layer_scale
if self.ls:
global init_value
print(f"Use layer_scale: {layer_scale}, init_values: {init_value}")
self.gamma_1 = nn.Parameter(
init_value * flow.ones((dim)), requires_grad=True
)
self.gamma_2 = nn.Parameter(
init_value * flow.ones((dim)), requires_grad=True
)
def forward(self, x):
x = x + self.pos_embed(x)
B, N, H, W = x.shape
x = x.flatten(2).transpose(1, 2)
if self.ls:
x = x + self.drop_path(self.gamma_1 * self.attn(self.norm1(x)))
x = x + self.drop_path(self.gamma_2 * self.mlp(self.norm2(x)))
else:
x = x + self.drop_path(self.attn(self.norm1(x)))
x = x + self.drop_path(self.mlp(self.norm2(x)))
x = x.transpose(1, 2).reshape(B, N, H, W)
return x
class head_embedding(nn.Module):
def __init__(self, in_channels, out_channels):
super(head_embedding, self).__init__()
self.proj = nn.Sequential(
nn.Conv2d(
in_channels,
out_channels // 2,
kernel_size=(3, 3),
stride=(2, 2),
padding=(1, 1),
),
nn.BatchNorm2d(out_channels // 2),
nn.GELU(),
nn.Conv2d(
out_channels // 2,
out_channels,
kernel_size=(3, 3),
stride=(2, 2),
padding=(1, 1),
),
nn.BatchNorm2d(out_channels),
)
def forward(self, x):
x = self.proj(x)
return x
class middle_embedding(nn.Module):
def __init__(self, in_channels, out_channels):
super(middle_embedding, self).__init__()
self.proj = nn.Sequential(
nn.Conv2d(
in_channels,
out_channels,
kernel_size=(3, 3),
stride=(2, 2),
padding=(1, 1),
),
nn.BatchNorm2d(out_channels),
)
def forward(self, x):
x = self.proj(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):
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])
self.img_size = img_size
self.patch_size = patch_size
self.num_patches = num_patches
self.norm = nn.LayerNorm(embed_dim)
self.proj = nn.Conv2d(
in_chans, embed_dim, kernel_size=patch_size, stride=patch_size
)
def forward(self, x):
B, C, H, W = x.shape
# FIXME look at relaxing size constraints
assert (
H == self.img_size[0] and W == self.img_size[1]
), f"Input image size ({H}*{W}) doesn't match model ({self.img_size[0]}*{self.img_size[1]})."
x = self.proj(x)
B, C, H, W = x.shape
x = x.flatten(2).transpose(1, 2)
x = self.norm(x)
x = x.reshape(B, H, W, -1).permute(0, 3, 1, 2).contiguous()
return x
class UniFormer(nn.Module):
""" UniFormer
A oneflow impl of : `UniFormer: Unified Transformer for Efficient Spatiotemporal Representation Learning` -
https://arxiv.org/abs/2201.04676
"""
def __init__(
self,
depth=[3, 4, 8, 3],
img_size=224,
in_chans=3,
num_classes=1000,
embed_dim=[64, 128, 320, 512],
head_dim=64,
mlp_ratio=4.0,
qkv_bias=True,
qk_scale=None,
representation_size=None,
drop_rate=0.0,
attn_drop_rate=0.0,
drop_path_rate=0.0,
norm_layer=None,
conv_stem=False,
):
"""
Args:
depth (list): depth of each stage
img_size (int, tuple): input image size
in_chans (int): number of input channels
num_classes (int): number of classes for classification head
embed_dim (list): embedding dimension of each stage
head_dim (int): head dimension
mlp_ratio (int): ratio of mlp hidden dim to embedding dim
qkv_bias (bool): enable bias for qkv if True
qk_scale (float): override default qk scale of head_dim ** -0.5 if set
representation_size (Optional[int]): enable and set representation layer (pre-logits) to this value if set
drop_rate (float): dropout rate
attn_drop_rate (float): attention dropout rate
drop_path_rate (float): stochastic depth rate
norm_layer (nn.Module): normalization layer
conv_stem (bool): whether to use overlapped patch stem
"""
super().__init__()
self.num_classes = num_classes
self.num_features = (
self.embed_dim
) = embed_dim # num_features for consistency with other models
norm_layer = norm_layer or partial(nn.LayerNorm, eps=1e-6)
if conv_stem:
self.patch_embed1 = head_embedding(
in_channels=in_chans, out_channels=embed_dim[0]
)
self.patch_embed2 = middle_embedding(
in_channels=embed_dim[0], out_channels=embed_dim[1]
)
self.patch_embed3 = middle_embedding(
in_channels=embed_dim[1], out_channels=embed_dim[2]
)
self.patch_embed4 = middle_embedding(
in_channels=embed_dim[2], out_channels=embed_dim[3]
)
else:
self.patch_embed1 = PatchEmbed(
img_size=img_size,
patch_size=4,
in_chans=in_chans,
embed_dim=embed_dim[0],
)
self.patch_embed2 = PatchEmbed(
img_size=img_size // 4,
patch_size=2,
in_chans=embed_dim[0],
embed_dim=embed_dim[1],
)
self.patch_embed3 = PatchEmbed(
img_size=img_size // 8,
patch_size=2,
in_chans=embed_dim[1],
embed_dim=embed_dim[2],
)
self.patch_embed4 = PatchEmbed(
img_size=img_size // 16,
patch_size=2,
in_chans=embed_dim[2],
embed_dim=embed_dim[3],
)
self.pos_drop = nn.Dropout(p=drop_rate)
dpr = [
x.item() for x in flow.linspace(0, drop_path_rate, sum(depth))
] # stochastic depth decay rule
num_heads = [dim // head_dim for dim in embed_dim]
self.blocks1 = nn.ModuleList(
[
CBlock(
dim=embed_dim[0],
num_heads=num_heads[0],
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[0])
]
)
self.blocks2 = nn.ModuleList(
[
CBlock(
dim=embed_dim[1],
num_heads=num_heads[1],
mlp_ratio=mlp_ratio,
qkv_bias=qkv_bias,
qk_scale=qk_scale,
drop=drop_rate,
attn_drop=attn_drop_rate,
drop_path=dpr[i + depth[0]],
norm_layer=norm_layer,
)
for i in range(depth[1])
]
)
self.blocks3 = nn.ModuleList(
[
SABlock(
dim=embed_dim[2],
num_heads=num_heads[2],
mlp_ratio=mlp_ratio,
qkv_bias=qkv_bias,
qk_scale=qk_scale,
drop=drop_rate,
attn_drop=attn_drop_rate,
drop_path=dpr[i + depth[0] + depth[1]],
norm_layer=norm_layer,
)
for i in range(depth[2])
]
)
self.blocks4 = nn.ModuleList(
[
SABlock(
dim=embed_dim[3],
num_heads=num_heads[3],
mlp_ratio=mlp_ratio,
qkv_bias=qkv_bias,
qk_scale=qk_scale,
drop=drop_rate,
attn_drop=attn_drop_rate,
drop_path=dpr[i + depth[0] + depth[1] + depth[2]],
norm_layer=norm_layer,
)
for i in range(depth[3])
]
)
self.norm = nn.BatchNorm2d(embed_dim[-1])
# Representation layer
if representation_size:
self.num_features = representation_size
self.pre_logits = nn.Sequential(
OrderedDict(
[
("fc", nn.Linear(embed_dim, representation_size)),
("act", nn.Tanh()),
]
)
)
else:
self.pre_logits = nn.Identity()
# Classifier head
self.head = (
nn.Linear(embed_dim[-1], num_classes) if num_classes > 0 else nn.Identity()
)
self.apply(self._init_weights)
def _init_weights(self, m):
if isinstance(m, nn.Linear):
trunc_normal_(m.weight, std=0.02)
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)
def no_weight_decay(self):
return {"pos_embed", "cls_token"}
def get_classifier(self):
return self.head
def reset_classifier(self, num_classes, global_pool=""):
self.num_classes = num_classes
self.head = (
nn.Linear(self.embed_dim, num_classes) if num_classes > 0 else nn.Identity()
)
def forward_features(self, x):
x = self.patch_embed1(x)
x = self.pos_drop(x)
for blk in self.blocks1:
x = blk(x)
x = self.patch_embed2(x)
for blk in self.blocks2:
x = blk(x)
x = self.patch_embed3(x)
for blk in self.blocks3:
x = blk(x)
x = self.patch_embed4(x)
for blk in self.blocks4:
x = blk(x)
x = self.norm(x)
x = self.pre_logits(x)
return x
def forward(self, x):
x = self.forward_features(x)
x = x.flatten(2).mean(-1)
x = self.head(x)
return x
def _create_uniformer(arch, pretrained=False, progress=True, **model_kwargs):
model = UniFormer(**model_kwargs)
if pretrained:
state_dict = load_state_dict_from_url(model_urls[arch], progress=progress)
model.load_state_dict(state_dict)
return model