|
""" Vision Transformer (ViT) in PyTorch |
|
|
|
A PyTorch implement of Vision Transformers as described in |
|
'An Image Is Worth 16 x 16 Words: Transformers for Image Recognition at Scale' - https://arxiv.org/abs/2010.11929 |
|
|
|
The official jax code is released and available at https://github.com/google-research/vision_transformer |
|
|
|
Status/TODO: |
|
* Models updated to be compatible with official impl. Args added to support backward compat for old PyTorch weights. |
|
* Weights ported from official jax impl for 384x384 base and small models, 16x16 and 32x32 patches. |
|
* Trained (supervised on ImageNet-1k) my custom 'small' patch model to 77.9, 'base' to 79.4 top-1 with this code. |
|
* Hopefully find time and GPUs for SSL or unsupervised pretraining on OpenImages w/ ImageNet fine-tune in future. |
|
|
|
Acknowledgments: |
|
* The paper authors for releasing code and weights, thanks! |
|
* I fixed my class token impl based on Phil Wang's https://github.com/lucidrains/vit-pytorch ... check it out |
|
for some einops/einsum fun |
|
* Simple transformer style inspired by Andrej Karpathy's https://github.com/karpathy/minGPT |
|
* Bert reference code checks against Huggingface Transformers and Tensorflow Bert |
|
|
|
Hacked together by / Copyright 2020 Ross Wightman |
|
""" |
|
|
|
from transformers import ( |
|
PreTrainedModel, |
|
PretrainedConfig, |
|
AutoConfig, |
|
AutoModel, |
|
AutoModelForImageClassification, |
|
) |
|
|
|
import math |
|
from functools import partial |
|
from itertools import repeat |
|
|
|
import torch |
|
import torch.nn as nn |
|
import torch.nn.functional as F |
|
from timm.data import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD |
|
|
|
TORCH_MAJOR = int(torch.__version__.split(".")[0]) |
|
TORCH_MINOR = int(torch.__version__.split(".")[1]) |
|
if TORCH_MAJOR == 1 and TORCH_MINOR < 8: |
|
from torch._six import container_abcs, int_classes |
|
else: |
|
import collections.abc as container_abcs |
|
|
|
int_classes = int |
|
|
|
|
|
|
|
def _ntuple(n): |
|
def parse(x): |
|
if isinstance(x, container_abcs.Iterable): |
|
return x |
|
return tuple(repeat(x, n)) |
|
|
|
return parse |
|
|
|
|
|
to_2tuple = _ntuple(2) |
|
|
|
|
|
def drop_path(x, drop_prob: float = 0.0, training: bool = False): |
|
"""Drop paths (Stochastic Depth) per sample (when applied in main path of residual blocks). |
|
|
|
This is the same as the DropConnect impl I created for EfficientNet, etc networks, however, |
|
the original name is misleading as 'Drop Connect' is a different form of dropout in a separate paper... |
|
See discussion: https://github.com/tensorflow/tpu/issues/494#issuecomment-532968956 ... I've opted for |
|
changing the layer and argument names to 'drop path' rather than mix DropConnect as a layer name and use |
|
'survival rate' as the argument. |
|
|
|
""" |
|
if drop_prob == 0.0 or not training: |
|
return x |
|
keep_prob = 1 - drop_prob |
|
shape = (x.shape[0],) + (1,) * ( |
|
x.ndim - 1 |
|
) |
|
random_tensor = keep_prob + torch.rand(shape, dtype=x.dtype, device=x.device) |
|
random_tensor.floor_() |
|
output = x.div(keep_prob) * random_tensor |
|
return output |
|
|
|
|
|
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 _cfg(url="", **kwargs): |
|
return { |
|
"url": url, |
|
"num_classes": 1000, |
|
"input_size": (3, 224, 224), |
|
"pool_size": None, |
|
"crop_pct": 0.9, |
|
"interpolation": "bicubic", |
|
"mean": IMAGENET_DEFAULT_MEAN, |
|
"std": IMAGENET_DEFAULT_STD, |
|
"first_conv": "patch_embed.proj", |
|
"classifier": "head", |
|
**kwargs, |
|
} |
|
|
|
|
|
default_cfgs = { |
|
|
|
"vit_small_patch16_224": _cfg( |
|
url="https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-weights/vit_small_p16_224-15ec54c9.pth", |
|
), |
|
"vit_base_patch16_224": _cfg( |
|
url="https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-vitjx/jx_vit_base_p16_224-80ecf9dd.pth", |
|
mean=(0.5, 0.5, 0.5), |
|
std=(0.5, 0.5, 0.5), |
|
), |
|
"vit_base_patch16_384": _cfg( |
|
url="https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-vitjx/jx_vit_base_p16_384-83fb41ba.pth", |
|
input_size=(3, 384, 384), |
|
mean=(0.5, 0.5, 0.5), |
|
std=(0.5, 0.5, 0.5), |
|
crop_pct=1.0, |
|
), |
|
"vit_base_patch32_384": _cfg( |
|
url="https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-vitjx/jx_vit_base_p32_384-830016f5.pth", |
|
input_size=(3, 384, 384), |
|
mean=(0.5, 0.5, 0.5), |
|
std=(0.5, 0.5, 0.5), |
|
crop_pct=1.0, |
|
), |
|
"vit_large_patch16_224": _cfg( |
|
url="https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-vitjx/jx_vit_large_p16_224-4ee7a4dc.pth", |
|
mean=(0.5, 0.5, 0.5), |
|
std=(0.5, 0.5, 0.5), |
|
), |
|
"vit_large_patch16_384": _cfg( |
|
url="https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-vitjx/jx_vit_large_p16_384-b3be5167.pth", |
|
input_size=(3, 384, 384), |
|
mean=(0.5, 0.5, 0.5), |
|
std=(0.5, 0.5, 0.5), |
|
crop_pct=1.0, |
|
), |
|
"vit_large_patch32_384": _cfg( |
|
url="https://github.com/rwightman/pytorch-image-models/releases/download/v0.1-vitjx/jx_vit_large_p32_384-9b920ba8.pth", |
|
input_size=(3, 384, 384), |
|
mean=(0.5, 0.5, 0.5), |
|
std=(0.5, 0.5, 0.5), |
|
crop_pct=1.0, |
|
), |
|
"vit_huge_patch16_224": _cfg(), |
|
"vit_huge_patch32_384": _cfg(input_size=(3, 384, 384)), |
|
|
|
"vit_small_resnet26d_224": _cfg(), |
|
"vit_small_resnet50d_s3_224": _cfg(), |
|
"vit_base_resnet26d_224": _cfg(), |
|
"vit_base_resnet50d_224": _cfg(), |
|
} |
|
|
|
|
|
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 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 |
|
|
|
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 Block(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.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, |
|
) |
|
|
|
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, |
|
) |
|
|
|
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): |
|
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.proj = nn.Conv2d( |
|
in_chans, embed_dim, kernel_size=patch_size, stride=patch_size |
|
) |
|
|
|
def forward(self, x): |
|
B, C, H, W = x.shape |
|
|
|
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).flatten(2).transpose(1, 2) |
|
return x |
|
|
|
|
|
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])) |
|
if isinstance(o, (list, tuple)): |
|
o = o[-1] |
|
feature_size = o.shape[-2:] |
|
feature_dim = o.shape[1] |
|
backbone.train(training) |
|
else: |
|
feature_size = to_2tuple(feature_size) |
|
if hasattr(self.backbone, "feature_info"): |
|
feature_dim = self.backbone.feature_info.channels()[-1] |
|
else: |
|
feature_dim = self.backbone.num_features |
|
self.num_patches = feature_size[0] * feature_size[1] |
|
self.proj = nn.Conv2d(feature_dim, embed_dim, 1) |
|
|
|
def forward(self, x): |
|
x = self.backbone(x) |
|
if isinstance(x, (list, tuple)): |
|
x = x[-1] |
|
x = self.proj(x).flatten(2).transpose(1, 2) |
|
return x |
|
|
|
|
|
class PatchEmbed_overlap(nn.Module): |
|
"""Image to Patch Embedding with overlapping patches""" |
|
|
|
def __init__( |
|
self, img_size=224, patch_size=16, stride_size=20, in_chans=3, embed_dim=768 |
|
): |
|
super().__init__() |
|
img_size = to_2tuple(img_size) |
|
patch_size = to_2tuple(patch_size) |
|
stride_size_tuple = to_2tuple(stride_size) |
|
self.num_x = (img_size[1] - patch_size[1]) // stride_size_tuple[1] + 1 |
|
self.num_y = (img_size[0] - patch_size[0]) // stride_size_tuple[0] + 1 |
|
print( |
|
"using stride: {}, and patch number is num_y{} * num_x{}".format( |
|
stride_size, self.num_y, self.num_x |
|
) |
|
) |
|
num_patches = self.num_x * self.num_y |
|
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=stride_size |
|
) |
|
for m in self.modules(): |
|
if isinstance(m, nn.Conv2d): |
|
n = m.kernel_size[0] * m.kernel_size[1] * m.out_channels |
|
m.weight.data.normal_(0, math.sqrt(2.0 / n)) |
|
elif isinstance(m, nn.BatchNorm2d): |
|
m.weight.data.fill_(1) |
|
m.bias.data.zero_() |
|
elif isinstance(m, nn.InstanceNorm2d): |
|
m.weight.data.fill_(1) |
|
m.bias.data.zero_() |
|
|
|
def forward(self, x): |
|
B, C, H, W = x.shape |
|
|
|
|
|
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) |
|
|
|
x = x.flatten(2).transpose(1, 2) |
|
return x |
|
|
|
|
|
class TransReID(nn.Module): |
|
"""Transformer-based Object Re-Identification""" |
|
|
|
@classmethod |
|
def from_config(cls, config): |
|
return cls( |
|
img_size=config.get("img_size", [384, 128]), |
|
patch_size=config.get("patch_size", 16), |
|
stride_size=config.get("stride_size", 16), |
|
in_chans=config.get("in_chans", 3), |
|
num_classes=config.get("num_classes", 1000), |
|
embed_dim=config.get("embed_dim", 768), |
|
depth=config.get("depth", 12), |
|
num_heads=config.get("num_heads", 12), |
|
mlp_ratio=config.get("mlp_ratio", 4.0), |
|
qkv_bias=config.get("qkv_bias", False), |
|
qk_scale=config.get("qk_scale", None), |
|
drop_rate=config.get("drop_rate", 0.0), |
|
attn_drop_rate=config.get("attn_drop_rate", 0.0), |
|
drop_path_rate=config.get("drop_path_rate", 0.0), |
|
camera=config.get("camera", 0), |
|
view=config.get("view", 0), |
|
local_feature=config.get("local_feature", False), |
|
sie_xishu=config.get("sie_xishu", 1.0), |
|
) |
|
|
|
def __init__( |
|
self, |
|
img_size=224, |
|
patch_size=16, |
|
stride_size=16, |
|
in_chans=3, |
|
num_classes=1000, |
|
embed_dim=768, |
|
depth=12, |
|
num_heads=12, |
|
mlp_ratio=4.0, |
|
qkv_bias=False, |
|
qk_scale=None, |
|
drop_rate=0.0, |
|
attn_drop_rate=0.0, |
|
camera=0, |
|
view=0, |
|
drop_path_rate=0.0, |
|
hybrid_backbone=None, |
|
norm_layer=nn.LayerNorm, |
|
local_feature=False, |
|
sie_xishu=1.0, |
|
): |
|
nn.Module.__init__(self) |
|
|
|
self.num_classes = num_classes |
|
self.num_features = self.embed_dim = ( |
|
embed_dim |
|
) |
|
self.local_feature = local_feature |
|
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_overlap( |
|
img_size=img_size, |
|
patch_size=patch_size, |
|
stride_size=stride_size, |
|
in_chans=in_chans, |
|
embed_dim=embed_dim, |
|
) |
|
|
|
num_patches = self.patch_embed.num_patches |
|
|
|
self.cls_token = nn.Parameter(torch.zeros(1, 1, embed_dim)) |
|
self.pos_embed = nn.Parameter(torch.zeros(1, num_patches + 1, embed_dim)) |
|
self.cam_num = camera |
|
self.view_num = view |
|
self.sie_xishu = sie_xishu |
|
|
|
if camera > 1 and view > 1: |
|
self.sie_embed = nn.Parameter(torch.zeros(camera * view, 1, embed_dim)) |
|
trunc_normal_(self.sie_embed, std=0.02) |
|
print( |
|
"camera number is : {} and viewpoint number is : {}".format( |
|
camera, view |
|
) |
|
) |
|
print("using SIE_Lambda is : {}".format(sie_xishu)) |
|
elif camera > 1: |
|
self.sie_embed = nn.Parameter(torch.zeros(camera, 1, embed_dim)) |
|
trunc_normal_(self.sie_embed, std=0.02) |
|
print("camera number is : {}".format(camera)) |
|
print("using SIE_Lambda is : {}".format(sie_xishu)) |
|
elif view > 1: |
|
self.sie_embed = nn.Parameter(torch.zeros(view, 1, embed_dim)) |
|
trunc_normal_(self.sie_embed, std=0.02) |
|
print("viewpoint number is : {}".format(view)) |
|
print("using SIE_Lambda is : {}".format(sie_xishu)) |
|
|
|
print("using drop_out rate is : {}".format(drop_rate)) |
|
print("using attn_drop_out rate is : {}".format(attn_drop_rate)) |
|
print("using drop_path rate is : {}".format(drop_path_rate)) |
|
|
|
self.pos_drop = nn.Dropout(p=drop_rate) |
|
dpr = [ |
|
x.item() for x in torch.linspace(0, drop_path_rate, depth) |
|
] |
|
|
|
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.norm = norm_layer(embed_dim) |
|
|
|
|
|
self.fc = ( |
|
nn.Linear(embed_dim, num_classes) if num_classes > 0 else nn.Identity() |
|
) |
|
trunc_normal_(self.cls_token, std=0.02) |
|
trunc_normal_(self.pos_embed, std=0.02) |
|
|
|
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) |
|
|
|
@torch.jit.ignore |
|
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.fc = ( |
|
nn.Linear(self.embed_dim, num_classes) if num_classes > 0 else nn.Identity() |
|
) |
|
|
|
def forward_features(self, x, camera_id, view_id): |
|
B = x.shape[0] |
|
x = self.patch_embed(x) |
|
|
|
cls_tokens = self.cls_token.expand( |
|
B, -1, -1 |
|
) |
|
x = torch.cat((cls_tokens, x), dim=1) |
|
|
|
if self.cam_num > 0 and self.view_num > 0: |
|
x = ( |
|
x |
|
+ self.pos_embed |
|
+ self.sie_xishu * self.sie_embed[camera_id * self.view_num + view_id] |
|
) |
|
elif self.cam_num > 0: |
|
x = x + self.pos_embed + self.sie_xishu * self.sie_embed[camera_id] |
|
elif self.view_num > 0: |
|
x = x + self.pos_embed + self.sie_xishu * self.sie_embed[view_id] |
|
else: |
|
x = x + self.pos_embed |
|
|
|
x = self.pos_drop(x) |
|
|
|
if self.local_feature: |
|
for blk in self.blocks: |
|
x = blk(x) |
|
|
|
x = self.norm(x) |
|
|
|
return x |
|
|
|
else: |
|
for blk in self.blocks: |
|
x = blk(x) |
|
|
|
x = self.norm(x) |
|
|
|
return x[:, 0] |
|
|
|
def forward(self, x, cam_label=None, view_label=None): |
|
x = self.forward_features(x, cam_label, view_label) |
|
return x |
|
|
|
def load_param(self, model_path): |
|
param_dict = torch.load(model_path, map_location="cpu") |
|
if "model" in param_dict: |
|
param_dict = param_dict["model"] |
|
if "state_dict" in param_dict: |
|
param_dict = param_dict["state_dict"] |
|
for k, v in param_dict.items(): |
|
|
|
if "head" in k or "dist" in k: |
|
continue |
|
if "patch_embed.proj.weight" in k and len(v.shape) < 4: |
|
|
|
O, I, H, W = self.patch_embed.proj.weight.shape |
|
v = v.reshape(O, -1, H, W) |
|
elif k == "pos_embed" and v.shape != self.pos_embed.shape: |
|
|
|
if "distilled" in model_path: |
|
print("distill need to choose right cls token in the pth") |
|
v = torch.cat([v[:, 0:1], v[:, 2:]], dim=1) |
|
v = resize_pos_embed( |
|
v, self.pos_embed, self.patch_embed.num_y, self.patch_embed.num_x |
|
) |
|
try: |
|
self.state_dict()[k].copy_(v) |
|
|
|
except: |
|
|
|
|
|
|
|
pass |
|
|
|
|
|
def resize_pos_embed(posemb, posemb_new, hight, width): |
|
|
|
|
|
ntok_new = posemb_new.shape[1] |
|
|
|
posemb_token, posemb_grid = posemb[:, :1], posemb[0, 1:] |
|
ntok_new -= 1 |
|
|
|
int(math.sqrt(len(posemb_grid))) |
|
print( |
|
"Resized position embedding from size:{} to size: {} with height:{} width: {}".format( |
|
posemb.shape, posemb_new.shape, hight, width |
|
) |
|
) |
|
posemb_grid = posemb_grid.reshape(1, 16, 8, -1).permute(0, 3, 1, 2) |
|
posemb_grid = F.interpolate(posemb_grid, size=(hight, width), mode="bilinear") |
|
posemb_grid = posemb_grid.permute(0, 2, 3, 1).reshape(1, hight * width, -1) |
|
posemb = torch.cat([posemb_token, posemb_grid], dim=1) |
|
return posemb |
|
|
|
|
|
def vit_base_patch16_224_TransReID( |
|
img_size=(256, 128), |
|
stride_size=16, |
|
drop_rate=0.0, |
|
attn_drop_rate=0.0, |
|
drop_path_rate=0.1, |
|
camera=0, |
|
view=0, |
|
local_feature=False, |
|
sie_xishu=1.5, |
|
**kwargs, |
|
): |
|
model = TransReID( |
|
img_size=img_size, |
|
patch_size=16, |
|
stride_size=stride_size, |
|
embed_dim=768, |
|
depth=12, |
|
num_heads=12, |
|
mlp_ratio=4, |
|
qkv_bias=True, |
|
camera=camera, |
|
view=view, |
|
drop_path_rate=drop_path_rate, |
|
drop_rate=drop_rate, |
|
attn_drop_rate=attn_drop_rate, |
|
norm_layer=partial(nn.LayerNorm, eps=1e-6), |
|
sie_xishu=sie_xishu, |
|
local_feature=local_feature, |
|
**kwargs, |
|
) |
|
|
|
return model |
|
|
|
|
|
def vit_small_patch16_224_TransReID( |
|
img_size=(256, 128), |
|
stride_size=16, |
|
drop_rate=0.0, |
|
attn_drop_rate=0.0, |
|
drop_path_rate=0.1, |
|
camera=0, |
|
view=0, |
|
local_feature=False, |
|
sie_xishu=1.5, |
|
**kwargs, |
|
): |
|
kwargs.setdefault("qk_scale", 768**-0.5) |
|
model = TransReID( |
|
img_size=img_size, |
|
patch_size=16, |
|
stride_size=stride_size, |
|
embed_dim=768, |
|
depth=8, |
|
num_heads=8, |
|
mlp_ratio=3.0, |
|
qkv_bias=False, |
|
drop_path_rate=drop_path_rate, |
|
camera=camera, |
|
view=view, |
|
drop_rate=drop_rate, |
|
attn_drop_rate=attn_drop_rate, |
|
norm_layer=partial(nn.LayerNorm, eps=1e-6), |
|
sie_xishu=sie_xishu, |
|
local_feature=local_feature, |
|
**kwargs, |
|
) |
|
|
|
return model |
|
|
|
|
|
def deit_small_patch16_224_TransReID( |
|
img_size=(256, 128), |
|
stride_size=16, |
|
drop_path_rate=0.1, |
|
drop_rate=0.0, |
|
attn_drop_rate=0.0, |
|
camera=0, |
|
view=0, |
|
local_feature=False, |
|
sie_xishu=1.5, |
|
**kwargs, |
|
): |
|
model = TransReID( |
|
img_size=img_size, |
|
patch_size=16, |
|
stride_size=stride_size, |
|
embed_dim=384, |
|
depth=12, |
|
num_heads=6, |
|
mlp_ratio=4, |
|
qkv_bias=True, |
|
drop_path_rate=drop_path_rate, |
|
drop_rate=drop_rate, |
|
attn_drop_rate=attn_drop_rate, |
|
camera=camera, |
|
view=view, |
|
sie_xishu=sie_xishu, |
|
local_feature=local_feature, |
|
norm_layer=partial(nn.LayerNorm, eps=1e-6), |
|
**kwargs, |
|
) |
|
|
|
return model |
|
|
|
|
|
def _no_grad_trunc_normal_(tensor, mean, std, a, b): |
|
|
|
|
|
def norm_cdf(x): |
|
|
|
return (1.0 + math.erf(x / math.sqrt(2.0))) / 2.0 |
|
|
|
if (mean < a - 2 * std) or (mean > b + 2 * std): |
|
print( |
|
"mean is more than 2 std from [a, b] in nn.init.trunc_normal_. " |
|
"The distribution of values may be incorrect.", |
|
) |
|
|
|
with torch.no_grad(): |
|
|
|
|
|
|
|
l = norm_cdf((a - mean) / std) |
|
u = norm_cdf((b - mean) / std) |
|
|
|
|
|
|
|
tensor.uniform_(2 * l - 1, 2 * u - 1) |
|
|
|
|
|
|
|
tensor.erfinv_() |
|
|
|
|
|
tensor.mul_(std * math.sqrt(2.0)) |
|
tensor.add_(mean) |
|
|
|
|
|
tensor.clamp_(min=a, max=b) |
|
return tensor |
|
|
|
|
|
def trunc_normal_(tensor, mean=0.0, std=1.0, a=-2.0, b=2.0): |
|
|
|
r"""Fills the input Tensor with values drawn from a truncated |
|
normal distribution. The values are effectively drawn from the |
|
normal distribution :math:`\mathcal{N}(\text{mean}, \text{std}^2)` |
|
with values outside :math:`[a, b]` redrawn until they are within |
|
the bounds. The method used for generating the random values works |
|
best when :math:`a \leq \text{mean} \leq b`. |
|
Args: |
|
tensor: an n-dimensional `torch.Tensor` |
|
mean: the mean of the normal distribution |
|
std: the standard deviation of the normal distribution |
|
a: the minimum cutoff value |
|
b: the maximum cutoff value |
|
Examples: |
|
>>> w = torch.empty(3, 5) |
|
>>> nn.init.trunc_normal_(w) |
|
""" |
|
return _no_grad_trunc_normal_(tensor, mean, std, a, b) |
|
|
|
|
|
class HAPTransReIDConfig(PretrainedConfig): |
|
model_type = "my-vit-b16" |
|
|
|
def __init__( |
|
self, |
|
img_size=[384, 128], |
|
stride_size=[16, 16], |
|
drop_rate=0.0, |
|
attn_drop_rate=0.0, |
|
drop_path_rate=0.1, |
|
camera=0, |
|
view=0, |
|
local_feature=True, |
|
sie_xishu=3.0, |
|
num_classes=-1, |
|
patch_size=16, |
|
in_chans=3, |
|
embed_dim=768, |
|
depth=12, |
|
num_heads=12, |
|
mlp_ratio=4.0, |
|
qkv_bias=False, |
|
qk_scale=None, |
|
hybrid_backbone=None, |
|
norm_layer_eps=1e-6, |
|
**kwargs, |
|
): |
|
|
|
super().__init__(**kwargs) |
|
|
|
self.img_size = img_size |
|
self.stride_size = stride_size |
|
self.drop_rate = drop_rate |
|
self.attn_drop_rate = attn_drop_rate |
|
self.drop_path_rate = drop_path_rate |
|
self.camera = camera |
|
self.view = view |
|
self.local_feature = local_feature |
|
self.sie_xishu = sie_xishu |
|
self.num_classes = num_classes |
|
self.patch_size = patch_size |
|
self.in_chans = in_chans |
|
self.embed_dim = embed_dim |
|
self.depth = depth |
|
self.num_heads = num_heads |
|
self.mlp_ratio = mlp_ratio |
|
self.qkv_bias = qkv_bias |
|
self.qk_scale = qk_scale |
|
self.hybrid_backbone = hybrid_backbone |
|
self.norm_layer_eps = norm_layer_eps |
|
|
|
|
|
|
|
|
|
class HAPTransReID(TransReID, PreTrainedModel): |
|
config_class = HAPTransReIDConfig |
|
|
|
def __init__(self, config): |
|
PreTrainedModel.__init__(self, config) |
|
self.config = config |
|
self.model = TransReID( |
|
img_size=config.img_size, |
|
stride_size=config.stride_size, |
|
drop_rate=config.drop_rate, |
|
attn_drop_rate=config.attn_drop_rate, |
|
drop_path_rate=config.drop_path_rate, |
|
camera=config.camera, |
|
view=config.view, |
|
local_feature=config.local_feature, |
|
sie_xishu=config.sie_xishu, |
|
num_classes=config.num_classes, |
|
patch_size=config.patch_size, |
|
in_chans=config.in_chans, |
|
embed_dim=config.embed_dim, |
|
depth=config.depth, |
|
num_heads=config.num_heads, |
|
mlp_ratio=config.mlp_ratio, |
|
qkv_bias=config.qkv_bias, |
|
qk_scale=config.qk_scale, |
|
norm_layer=partial(nn.LayerNorm, eps=config.norm_layer_eps), |
|
) |
|
self.model.hidden_size = self.model.vision_width = config.embed_dim |
|
def forward(self, x): |
|
return self.model(x, cam_label=None, view_label=None) |
|
|
|
@classmethod |
|
def from_config(cls, config={}, from_path=None, from_pretrained=None): |
|
''' |
|
vision_width = hidden_size = 768, just for get information |
|
not used in the model |
|
''' |
|
model = vit_base_patch16_224_TransReID( |
|
img_size=config.get("img_size", [384, 128]), |
|
stride_size=config.get("stride_size", [16, 16]), |
|
drop_rate=config.get("drop_rate", 0.0), |
|
attn_drop_rate=config.get("attn_drop_rate", 0.0), |
|
drop_path_rate=config.get("drop_path_rate", 0.1), |
|
camera=config.get("camera", 0), |
|
view=config.get("view", 0), |
|
local_feature=config.get("local_feature", True), |
|
sie_xishu=config.get("sie_xishu", 3.0), |
|
num_classes=config.get("num_classes", -1), |
|
|
|
|
|
) |
|
model.vision_width = model.hidden_size = 768 |
|
|
|
if from_path is not None: |
|
model.load_param(from_path) |
|
|
|
return model |
|
|