Spaces:
Running
Running
File size: 7,078 Bytes
2ba4412 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 |
import os
import torch
import logging
import open_clip
import numpy as np
import torch.nn as nn
import torchvision.transforms as T
from utils.registry_class import EMBEDDER
@EMBEDDER.register_class()
class FrozenOpenCLIPEmbedder(nn.Module):
"""
Uses the OpenCLIP transformer encoder for text
"""
LAYERS = [
#"pooled",
"last",
"penultimate"
]
def __init__(self, pretrained, arch="ViT-H-14", device="cuda", max_length=77,
freeze=True, layer="last"):
super().__init__()
assert layer in self.LAYERS
model, _, _ = open_clip.create_model_and_transforms(arch, device=torch.device('cpu'), pretrained=pretrained)
del model.visual
self.model = model
self.device = device
self.max_length = max_length
if freeze:
self.freeze()
self.layer = layer
if self.layer == "last":
self.layer_idx = 0
elif self.layer == "penultimate":
self.layer_idx = 1
else:
raise NotImplementedError()
def freeze(self):
self.model = self.model.eval()
for param in self.parameters():
param.requires_grad = False
def forward(self, text):
tokens = open_clip.tokenize(text)
z = self.encode_with_transformer(tokens.to(self.device))
return z
def encode_with_transformer(self, text):
x = self.model.token_embedding(text) # [batch_size, n_ctx, d_model]
x = x + self.model.positional_embedding
x = x.permute(1, 0, 2) # NLD -> LND
x = self.text_transformer_forward(x, attn_mask=self.model.attn_mask)
x = x.permute(1, 0, 2) # LND -> NLD
x = self.model.ln_final(x)
return x
def text_transformer_forward(self, x: torch.Tensor, attn_mask = None):
for i, r in enumerate(self.model.transformer.resblocks):
if i == len(self.model.transformer.resblocks) - self.layer_idx:
break
if self.model.transformer.grad_checkpointing and not torch.jit.is_scripting():
x = checkpoint(r, x, attn_mask)
else:
x = r(x, attn_mask=attn_mask)
return x
def encode(self, text):
return self(text)
@EMBEDDER.register_class()
class FrozenOpenCLIPVisualEmbedder(nn.Module):
"""
Uses the OpenCLIP transformer encoder for text
"""
LAYERS = [
#"pooled",
"last",
"penultimate"
]
def __init__(self, pretrained, vit_resolution=(224, 224), arch="ViT-H-14", device="cuda", max_length=77,
freeze=True, layer="last"):
super().__init__()
assert layer in self.LAYERS
model, _, preprocess = open_clip.create_model_and_transforms(
arch, device=torch.device('cpu'), pretrained=pretrained)
del model.transformer
self.model = model
data_white = np.ones((vit_resolution[0], vit_resolution[1], 3), dtype=np.uint8)*255
self.white_image = preprocess(T.ToPILImage()(data_white)).unsqueeze(0)
self.device = device
self.max_length = max_length # 77
if freeze:
self.freeze()
self.layer = layer # 'penultimate'
if self.layer == "last":
self.layer_idx = 0
elif self.layer == "penultimate":
self.layer_idx = 1
else:
raise NotImplementedError()
def freeze(self):
self.model = self.model.eval()
for param in self.parameters():
param.requires_grad = False
def forward(self, image):
# tokens = open_clip.tokenize(text)
z = self.model.encode_image(image.to(self.device))
return z
def encode_with_transformer(self, text):
x = self.model.token_embedding(text) # [batch_size, n_ctx, d_model]
x = x + self.model.positional_embedding
x = x.permute(1, 0, 2) # NLD -> LND
x = self.text_transformer_forward(x, attn_mask=self.model.attn_mask)
x = x.permute(1, 0, 2) # LND -> NLD
x = self.model.ln_final(x)
return x
def text_transformer_forward(self, x: torch.Tensor, attn_mask = None):
for i, r in enumerate(self.model.transformer.resblocks):
if i == len(self.model.transformer.resblocks) - self.layer_idx:
break
if self.model.transformer.grad_checkpointing and not torch.jit.is_scripting():
x = checkpoint(r, x, attn_mask)
else:
x = r(x, attn_mask=attn_mask)
return x
def encode(self, text):
return self(text)
@EMBEDDER.register_class()
class FrozenOpenCLIPTextVisualEmbedder(nn.Module):
"""
Uses the OpenCLIP transformer encoder for text
"""
LAYERS = [
#"pooled",
"last",
"penultimate"
]
def __init__(self, pretrained, arch="ViT-H-14", device="cuda", max_length=77,
freeze=True, layer="last", **kwargs):
super().__init__()
assert layer in self.LAYERS
model, _, _ = open_clip.create_model_and_transforms(arch, device=torch.device('cpu'), pretrained=pretrained)
self.model = model
self.device = device
self.max_length = max_length
if freeze:
self.freeze()
self.layer = layer
if self.layer == "last":
self.layer_idx = 0
elif self.layer == "penultimate":
self.layer_idx = 1
else:
raise NotImplementedError()
def freeze(self):
self.model = self.model.eval()
for param in self.parameters():
param.requires_grad = False
def forward(self, image=None, text=None):
xi = self.model.encode_image(image.to(self.device)) if image is not None else None
tokens = open_clip.tokenize(text)
xt, x = self.encode_with_transformer(tokens.to(self.device))
return xi, xt, x
def encode_with_transformer(self, text):
x = self.model.token_embedding(text) # [batch_size, n_ctx, d_model]
x = x + self.model.positional_embedding
x = x.permute(1, 0, 2) # NLD -> LND
x = self.text_transformer_forward(x, attn_mask=self.model.attn_mask)
x = x.permute(1, 0, 2) # LND -> NLD
x = self.model.ln_final(x)
xt = x[torch.arange(x.shape[0]), text.argmax(dim=-1)] @ self.model.text_projection
return xt, x
def encode_image(self, image):
return self.model.visual(image)
def text_transformer_forward(self, x: torch.Tensor, attn_mask = None):
for i, r in enumerate(self.model.transformer.resblocks):
if i == len(self.model.transformer.resblocks) - self.layer_idx:
break
if self.model.transformer.grad_checkpointing and not torch.jit.is_scripting():
x = checkpoint(r, x, attn_mask)
else:
x = r(x, attn_mask=attn_mask)
return x
def encode(self, text):
return self(text)
|