File size: 10,793 Bytes
df66507 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 | import torch
import os
import torch.nn as nn
from clip import clip
from clip.simple_tokenizer import SimpleTokenizer as _Tokenizer
_tokenizer = _Tokenizer()
def cuda(tensor, is_cuda):
if is_cuda:
return tensor.cuda()
else:
return tensor
class Hook:
def __init__(self, name, module):
self.name = name
self.hook = module.register_forward_hook(self.hook_fn)
def hook_fn(self, module, input, output):
self.input = input
self.output = output
def close(self):
self.hook.remove()
# Load the CLIP model
def load_clip_to_cpu(cfg):
backbone_name = cfg.model_name
url = clip._MODELS[backbone_name]
model_path = clip._download(url, os.path.expanduser("~/.cache/clip"))
try:
# Loading JIT archive
model = torch.jit.load(model_path, map_location="cpu").eval()
state_dict = None
except RuntimeError:
state_dict = torch.load(model_path, map_location="cpu")
model = clip.build_model(state_dict or model.state_dict())
return model
# Text encoder that maps prompt tokens into feature vectors
class TextEncoder(nn.Module):
def __init__(self, clip_model):
super().__init__()
self.transformer = clip_model.transformer
self.positional_embedding = clip_model.positional_embedding
self.ln_final = clip_model.ln_final
self.text_projection = clip_model.text_projection
self.dtype = clip_model.dtype
def forward(self, prompts, tokenized_prompts):
"""
:param prompts: Tensor representation of text prompts
:param tokenized_prompts: Tokenized text prompts
:return: Text features
"""
# Add prompts and positional embeddings
x = prompts + self.positional_embedding.type(self.dtype)
x = x.permute(1, 0, 2) # NLD -> LND
x = self.transformer(x)
x = x.permute(1, 0, 2) # LND -> NLD
x = self.ln_final(x).type(self.dtype)
# x.shape = [batch_size, n_ctx, transformer.width]
# Take features from the EOT embedding (EOT token is the highest number in each sequence)
x = x[torch.arange(x.shape[0]), tokenized_prompts.argmax(dim=-1)] @ self.text_projection
return x
# Learn and generate context-aware prompts
class PromptLearner(nn.Module):
def __init__(self, cfg, classnames, clip_model):
super().__init__()
n_cls = len(classnames)
n_ctx = cfg.n_ctx
ctx_init = cfg.ctx_init
dtype = clip_model.dtype
ctx_dim = clip_model.ln_final.weight.shape[0]
clip_imsize = clip_model.visual.input_resolution
cfg_imsize = cfg.img_size
assert cfg_imsize == clip_imsize, f"cfg_imsize ({cfg_imsize}) must equal to clip_imsize ({clip_imsize})"
if ctx_init:
# Use CLIP tokenize to determine the actual number of tokens
# instead of counting spaces manually
ctx_init = ctx_init.replace("_", " ")
with torch.no_grad():
tok = clip.tokenize(ctx_init) # [1, L]
eot = tok.argmax(dim=-1).item() # Index of the final EOT token
n_ctx = eot - 1 # Remove [SOS] and keep only context tokens
emb = clip_model.token_embedding(tok).to(torch.float32) # Use FP32 for parameters, not fp16
# Extract the token vectors corresponding to the context
# remove [SOS] and keep n_ctx tokens
ctx_vectors = emb[0, 1:1 + n_ctx, :].contiguous() # [n_ctx, D]
prompt_prefix = ctx_init
else:
# Random initialization: parameters must be FP32
if cfg.csc:
print("Initializing class-specific contexts")
ctx_vectors = torch.empty(n_cls, n_ctx, ctx_dim, dtype=torch.float32)
else:
print("Initializing a generic context")
ctx_vectors = torch.empty(n_ctx, ctx_dim, dtype=torch.float32)
nn.init.normal_(ctx_vectors, std=0.02)
prompt_prefix = " ".join(["X"] * n_ctx)
print(f'Initial context: "{prompt_prefix}"')
print(f"Number of context words (tokens): {n_ctx}")
self.ctx = nn.Parameter(ctx_vectors) # To be optimized
classnames = [name.replace("_", " ") for name in classnames]
name_lens = [len(_tokenizer.encode(name)) for name in classnames]
prompts = [prompt_prefix + " " + name + "." for name in classnames]
tokenized_prompts = torch.cat([clip.tokenize(p) for p in prompts])
with torch.no_grad():
embedding = clip_model.token_embedding(tokenized_prompts).type(dtype)
# These token vectors will be saved in save_model(),
# but they should be ignored in load_model() because we want to use
# those computed using the current class names
self.register_buffer("token_prefix", embedding[:, :1, :]) # SOS
self.register_buffer("token_suffix", embedding[:, 1 + n_ctx:, :]) # CLS, EOS
self.n_cls = n_cls
self.n_ctx = n_ctx
self.tokenized_prompts = tokenized_prompts # torch.Tensor
self.name_lens = name_lens
self.class_token_position = cfg.class_token_position
def forward(self):
ctx = self.ctx
if ctx.dim() == 2:
ctx = ctx.unsqueeze(0).expand(self.n_cls, -1, -1)
ctx = ctx.to(self.token_prefix.dtype)
prefix = self.token_prefix
suffix = self.token_suffix
if self.class_token_position == "end":
prompts = torch.cat(
[
prefix, # (n_cls, 1, dim)
ctx, # (n_cls, n_ctx, dim)
suffix, # (n_cls, *, dim)
],
dim=1,
)
elif self.class_token_position == "middle":
half_n_ctx = self.n_ctx // 2
prompts = []
for i in range(self.n_cls):
name_len = self.name_lens[i]
prefix_i = prefix[i: i + 1, :, :]
class_i = suffix[i: i + 1, :name_len, :]
suffix_i = suffix[i: i + 1, name_len:, :]
ctx_i_half1 = ctx[i: i + 1, :half_n_ctx, :]
ctx_i_half2 = ctx[i: i + 1, half_n_ctx:, :]
prompt = torch.cat(
[
prefix_i, # (1, 1, dim)
ctx_i_half1, # (1, n_ctx//2, dim)
class_i, # (1, name_len, dim)
ctx_i_half2, # (1, n_ctx//2, dim)
suffix_i, # (1, *, dim)
],
dim=1,
)
prompts.append(prompt)
prompts = torch.cat(prompts, dim=0)
elif self.class_token_position == "front":
prompts = []
for i in range(self.n_cls):
name_len = self.name_lens[i]
prefix_i = prefix[i: i + 1, :, :]
class_i = suffix[i: i + 1, :name_len, :]
suffix_i = suffix[i: i + 1, name_len:, :]
ctx_i = ctx[i: i + 1, :, :]
prompt = torch.cat(
[
prefix_i, # (1, 1, dim)
class_i, # (1, name_len, dim)
ctx_i, # (1, n_ctx, dim)
suffix_i, # (1, *, dim)
],
dim=1,
)
prompts.append(prompt)
prompts = torch.cat(prompts, dim=0)
else:
raise ValueError
return prompts
# Combine the image encoder and the text encoder
class DoUNet(nn.Module):
def __init__(self, cfg, clip_model, proj_dim=1024):
super().__init__()
classnames = cfg.classnames
self.clip = clip_model
self.image_encoder = clip_model.visual
self.prompt_learner = PromptLearner(cfg, classnames, clip_model)
self.tokenized_prompts = self.prompt_learner.tokenized_prompts
self.text_encoder = TextEncoder(clip_model)
self.logit_scale = clip_model.logit_scale
self.dtype = clip_model.dtype
# Register hooks to capture intermediate layer outputs
hook_candidates = [(name, module) for name, module in self.clip.visual.named_modules() if "ln_2" in name]
hook_candidates.sort(key=lambda x: x[0]) # Ensure stable layer ordering
self.hooks = [Hook(name, module) for name, module in hook_candidates]
H = len(self.hooks)
self.alpha = nn.Parameter(torch.randn(1, H, proj_dim)) # [1, H, D]
self.layer_mu = nn.ModuleList([
nn.Linear(proj_dim, proj_dim) for _ in range(H)
])
self.layer_logvar = nn.ModuleList([
nn.Linear(proj_dim, proj_dim) for _ in range(H)
])
def reparameterize(self, mu, logvar):
std = torch.exp(0.5 * logvar) # logvar = log(σ^2)
eps = torch.randn_like(std) # Same shape/device/dtype as std
return mu + std * eps
def forward(self, image, depth=None, return_features=False):
with torch.no_grad():
_ = self.clip.encode_image(image)
cls_list = [h.output[0, :, :].to(torch.float32) for h in self.hooks]
g = torch.stack(cls_list, dim=1).float()
mu_layers = []
logvar_layers = []
z_layers = []
for idx, feat in enumerate(torch.unbind(g, dim=1)):
mu_i = self.layer_mu[idx](feat)
logvar_i = self.layer_logvar[idx](feat)
z_i = self.reparameterize(mu_i, logvar_i)
mu_layers.append(mu_i)
logvar_layers.append(logvar_i)
z_layers.append(z_i)
mu = torch.stack(mu_layers, dim=1)
logvar = torch.stack(logvar_layers, dim=1)
z_stack = torch.stack(z_layers, dim=1)
for h in self.hooks:
h.output = None # Optional: clear cache
w = torch.softmax(self.alpha, dim=1).to(z_stack.dtype)
if w.size(0) != z_stack.size(0):
w = w.expand(z_stack.size(0), -1, -1)
z = (w * z_stack).sum(dim=1)
z = z.to(self.dtype)
image_features = z @ self.clip.visual.proj
prompts = self.prompt_learner()
tokenized_prompts = self.tokenized_prompts
text_features = self.text_encoder(prompts, tokenized_prompts)
image_features = image_features / image_features.norm(dim=-1, keepdim=True)
text_features = text_features / text_features.norm(dim=-1, keepdim=True)
logit_scale = self.logit_scale.exp()
logits = logit_scale * image_features @ text_features.t()
if return_features:
return logits, (mu, logvar), image_features, text_features, z_stack
else:
return logits
|