310 lines
9.9 KiB
Python
310 lines
9.9 KiB
Python
# These codes are copied from modelscope revision c58451baead80d83281f063d12fb377fad415257
|
|
# The implementation is adopted from OpenAI-CLIP,
|
|
# made pubicly available under the MIT License at https://github.com/openai/CLIP
|
|
|
|
import math
|
|
import sys
|
|
from collections import OrderedDict
|
|
from functools import reduce
|
|
from operator import mul
|
|
|
|
import numpy as np
|
|
import torch
|
|
import torch.nn as nn
|
|
import torch.nn.functional as F
|
|
from PIL import Image
|
|
from torchvision import models
|
|
|
|
from .utils import convert_weights, load_pretrained
|
|
|
|
|
|
class Bottleneck(nn.Module):
|
|
expansion = 4
|
|
|
|
def __init__(self, inplanes, planes, stride=1):
|
|
super().__init__()
|
|
|
|
# all conv layers have stride 1. an avgpool is performed
|
|
# after the second convolution when stride > 1
|
|
self.conv1 = nn.Conv2d(inplanes, planes, 1, bias=False)
|
|
self.bn1 = nn.BatchNorm2d(planes)
|
|
|
|
self.conv2 = nn.Conv2d(planes, planes, 3, padding=1, bias=False)
|
|
self.bn2 = nn.BatchNorm2d(planes)
|
|
|
|
self.avgpool = nn.AvgPool2d(stride) if stride > 1 else nn.Identity()
|
|
|
|
self.conv3 = nn.Conv2d(planes, planes * self.expansion, 1, bias=False)
|
|
self.bn3 = nn.BatchNorm2d(planes * self.expansion)
|
|
|
|
self.relu = nn.ReLU(inplace=True)
|
|
self.downsample = None
|
|
self.stride = stride
|
|
|
|
if stride > 1 or inplanes != planes * Bottleneck.expansion:
|
|
# downsampling layer is prepended with an avgpool,
|
|
# and the subsequent convolution has stride 1
|
|
self.downsample = nn.Sequential(
|
|
OrderedDict([('-1', nn.AvgPool2d(stride)),
|
|
('0',
|
|
nn.Conv2d(
|
|
inplanes,
|
|
planes * self.expansion,
|
|
1,
|
|
stride=1,
|
|
bias=False)),
|
|
('1', nn.BatchNorm2d(planes * self.expansion))]))
|
|
|
|
def forward(self, x: torch.Tensor):
|
|
identity = x
|
|
|
|
out = self.relu(self.bn1(self.conv1(x)))
|
|
out = self.relu(self.bn2(self.conv2(out)))
|
|
out = self.avgpool(out)
|
|
out = self.bn3(self.conv3(out))
|
|
|
|
if self.downsample is not None:
|
|
identity = self.downsample(x)
|
|
|
|
out += identity
|
|
out = self.relu(out)
|
|
return out
|
|
|
|
|
|
class AttentionPool2d(nn.Module):
|
|
|
|
def __init__(self,
|
|
spacial_dim: int,
|
|
embed_dim: int,
|
|
num_heads: int,
|
|
output_dim: int = None):
|
|
super().__init__()
|
|
self.positional_embedding = nn.Parameter(
|
|
torch.randn(spacial_dim**2 + 1, embed_dim) / embed_dim**0.5)
|
|
self.k_proj = nn.Linear(embed_dim, embed_dim)
|
|
self.q_proj = nn.Linear(embed_dim, embed_dim)
|
|
self.v_proj = nn.Linear(embed_dim, embed_dim)
|
|
self.c_proj = nn.Linear(embed_dim, output_dim or embed_dim)
|
|
self.num_heads = num_heads
|
|
|
|
def forward(self, x):
|
|
x = x.reshape(x.shape[0], x.shape[1],
|
|
x.shape[2] * x.shape[3]).permute(2, 0, 1)
|
|
x = torch.cat([x.mean(dim=0, keepdim=True), x], dim=0)
|
|
x = x + self.positional_embedding[:, None, :].to(x.dtype)
|
|
x, _ = F.multi_head_attention_forward(
|
|
query=x,
|
|
key=x,
|
|
value=x,
|
|
embed_dim_to_check=x.shape[-1],
|
|
num_heads=self.num_heads,
|
|
q_proj_weight=self.q_proj.weight,
|
|
k_proj_weight=self.k_proj.weight,
|
|
v_proj_weight=self.v_proj.weight,
|
|
in_proj_weight=None,
|
|
in_proj_bias=torch.cat(
|
|
[self.q_proj.bias, self.k_proj.bias, self.v_proj.bias]),
|
|
bias_k=None,
|
|
bias_v=None,
|
|
add_zero_attn=False,
|
|
dropout_p=0,
|
|
out_proj_weight=self.c_proj.weight,
|
|
out_proj_bias=self.c_proj.bias,
|
|
use_separate_proj_weight=True,
|
|
training=self.training,
|
|
need_weights=False)
|
|
|
|
return x[0]
|
|
|
|
|
|
class LayerNorm(nn.LayerNorm):
|
|
"""Subclass torch's LayerNorm to handle fp16."""
|
|
|
|
def forward(self, x: torch.Tensor):
|
|
orig_type = x.dtype
|
|
ret = super().forward(x.type(torch.float32))
|
|
return ret.type(orig_type)
|
|
|
|
|
|
class QuickGELU(nn.Module):
|
|
|
|
def forward(self, x: torch.Tensor):
|
|
return x * torch.sigmoid(1.702 * x)
|
|
|
|
|
|
class ResidualAttentionBlock(nn.Module):
|
|
|
|
def __init__(self,
|
|
d_model: int,
|
|
n_head: int,
|
|
attn_mask: torch.Tensor = None):
|
|
super().__init__()
|
|
|
|
self.attn = nn.MultiheadAttention(d_model, n_head)
|
|
self.ln_1 = LayerNorm(d_model)
|
|
self.mlp = nn.Sequential(
|
|
OrderedDict([('c_fc', nn.Linear(d_model, d_model * 4)),
|
|
('gelu', QuickGELU()),
|
|
('c_proj', nn.Linear(d_model * 4, d_model))]))
|
|
self.ln_2 = LayerNorm(d_model)
|
|
self.attn_mask = attn_mask
|
|
|
|
def attention(self, x: torch.Tensor):
|
|
self.attn_mask = self.attn_mask.to(
|
|
dtype=x.dtype,
|
|
device=x.device) if self.attn_mask is not None else None
|
|
return self.attn(
|
|
x, x, x, need_weights=False, attn_mask=self.attn_mask)[0]
|
|
|
|
def forward(self, x: torch.Tensor, idx):
|
|
features = {}
|
|
x_norm = self.ln_1(x)
|
|
features['layer_{}_pre_attn'.format(idx)] = x_norm.permute(1, 0, 2)
|
|
attn = self.attention(x_norm)
|
|
features['layer_{}_attn'.format(idx)] = attn.permute(1, 0, 2)
|
|
x = x + attn
|
|
mlp = self.mlp(self.ln_2(x))
|
|
features['layer_{}_mlp'.format(idx)] = mlp.permute(1, 0, 2)
|
|
x = x + mlp
|
|
return x, features
|
|
|
|
|
|
class Transformer(nn.Module):
|
|
|
|
def __init__(self,
|
|
width: int,
|
|
layers: int,
|
|
heads: int,
|
|
attn_mask: torch.Tensor = None):
|
|
super().__init__()
|
|
self.width = width
|
|
self.layers = layers
|
|
self.resblocks = nn.ModuleList()
|
|
for i in range(layers):
|
|
block = ResidualAttentionBlock(width, heads, attn_mask)
|
|
self.resblocks.append(block)
|
|
|
|
def forward(self, x: torch.Tensor):
|
|
features = {}
|
|
for idx, block in enumerate(self.resblocks):
|
|
x, block_feats = block(x, idx)
|
|
features.update(block_feats)
|
|
return x, features
|
|
|
|
|
|
class VisualTransformer(nn.Module):
|
|
|
|
def __init__(self, input_resolution: int, patch_size: int, width: int,
|
|
layers: int, heads: int, output_dim: int):
|
|
super().__init__()
|
|
print(input_resolution, patch_size, width, layers, heads, output_dim)
|
|
self.input_resolution = input_resolution
|
|
self.output_dim = output_dim
|
|
self.conv1 = nn.Conv2d(
|
|
in_channels=3,
|
|
out_channels=width,
|
|
kernel_size=patch_size,
|
|
stride=patch_size,
|
|
bias=False)
|
|
|
|
scale = width**-0.5
|
|
self.class_embedding = nn.Parameter(scale * torch.randn(width))
|
|
self.positional_embedding = nn.Parameter(scale * torch.randn(
|
|
(input_resolution // patch_size)**2 + 1, width))
|
|
self.ln_pre = LayerNorm(width)
|
|
|
|
self.transformer = Transformer(width, layers, heads)
|
|
|
|
self.ln_post = LayerNorm(width)
|
|
self.proj = nn.Parameter(scale * torch.randn(width, output_dim))
|
|
|
|
def forward(self, x: torch.Tensor, return_all=True):
|
|
x = self.conv1(x) # shape = [*, width, grid, grid]
|
|
x = x.reshape(x.shape[0], x.shape[1],
|
|
-1) # shape = [*, width, grid ** 2]
|
|
x = x.permute(0, 2, 1) # shape = [*, grid ** 2, width]
|
|
zeros = torch.zeros(
|
|
x.shape[0], 1, x.shape[-1], dtype=x.dtype, device=x.device)
|
|
# shape = [*, grid ** 2 + 1, width]
|
|
x = torch.cat([self.class_embedding.to(x.dtype) + zeros, x], dim=1)
|
|
x = x + self.positional_embedding.to(x.dtype)
|
|
x = self.ln_pre(x)
|
|
|
|
x = x.permute(1, 0, 2) # NLD -> LND
|
|
x, features = self.transformer(x)
|
|
x = x.permute(1, 0, 2) # LND -> NLD
|
|
|
|
x = self.ln_post(x[:, 0, :])
|
|
|
|
if return_all:
|
|
features['pre_logits'] = x
|
|
return features
|
|
|
|
if self.proj is not None:
|
|
x = x @ self.proj
|
|
|
|
return x
|
|
|
|
|
|
class CLIPNet(nn.Module):
|
|
|
|
def __init__(self, arch_name, pretrained, **kwargs):
|
|
super(CLIPNet, self).__init__()
|
|
|
|
if arch_name == 'CLIP_ViTB32':
|
|
self.clip = VisualTransformer(
|
|
input_resolution=224,
|
|
patch_size=32,
|
|
width=768,
|
|
layers=12,
|
|
heads=12,
|
|
output_dim=512)
|
|
|
|
elif arch_name in ('CLIP_ViTB16', 'CLIP_ViTB16_FP16'):
|
|
self.clip = VisualTransformer(
|
|
input_resolution=224,
|
|
patch_size=16,
|
|
width=768,
|
|
layers=12,
|
|
heads=12,
|
|
output_dim=512)
|
|
|
|
elif arch_name in ('CLIP_ViTL14', 'CLIP_ViTL14_FP16'):
|
|
self.clip = VisualTransformer(
|
|
input_resolution=224,
|
|
patch_size=14,
|
|
width=1024,
|
|
layers=24,
|
|
heads=16,
|
|
output_dim=768)
|
|
|
|
else:
|
|
raise KeyError(f'Unsupported arch_name for CLIP, {arch_name}')
|
|
|
|
def forward(self, input_data):
|
|
output = self.clip(input_data)
|
|
return output
|
|
|
|
|
|
def CLIP(arch_name='CLIP_RN50',
|
|
use_pretrain=False,
|
|
load_from='',
|
|
state_dict=None,
|
|
**kwargs):
|
|
model = CLIPNet(arch_name=arch_name, pretrained=None, **kwargs)
|
|
if use_pretrain:
|
|
if arch_name.endswith('FP16'):
|
|
convert_weights(model.clip)
|
|
load_pretrained(model.clip, state_dict, load_from)
|
|
return model
|
|
|
|
|
|
class ProbingModel(torch.nn.Module):
|
|
|
|
def __init__(self, feat_size, num_classes):
|
|
super(ProbingModel, self).__init__()
|
|
self.linear = torch.nn.Linear(feat_size, num_classes)
|
|
|
|
def forward(self, x):
|
|
return self.linear(x)
|