File size: 8,027 Bytes
d992912 | 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 | """
engine/encoder.py
FashionCLIPEncoder — wraps a HuggingFace CLIP model for text and image encoding.
Extracted from finalized_search_engine_full_script.py (lines 482-652).
"""
import logging
from pathlib import Path
from typing import List, Optional
import numpy as np
import torch
import torch.nn.functional as F
from PIL import Image
from transformers import CLIPModel, CLIPProcessor
from backend.app.config import SearchConfig
logger = logging.getLogger(__name__)
__all__ = ["FashionCLIPEncoder"]
class FashionCLIPEncoder:
"""
v3.1 — Handles models that return BaseModelOutputWithPooling
instead of raw tensors from get_text_features / get_image_features.
"""
def __init__(self, config: SearchConfig):
self.config = config
self.device = config.device
self.model = None
self.processor = None
self.model_name = None
self._load_model()
def _load_model(self):
models_to_try = [self.config.primary_model, self.config.fallback_model]
for model_name in models_to_try:
try:
logger.info(f"Loading model: {model_name}")
kwargs = {}
if self.config.hf_token:
kwargs['token'] = self.config.hf_token
self.model = CLIPModel.from_pretrained(model_name, **kwargs)
self.processor = CLIPProcessor.from_pretrained(model_name, **kwargs)
self.model = self.model.to(self.device)
self.model.eval()
self.model_name = model_name
# ── Probe the model to find actual embedding dim ──
test_inputs = self.processor(
text=["test"], return_tensors="pt",
padding=True, truncation=True, max_length=77,
)
test_inputs = {k: v.to(self.device) for k, v in test_inputs.items()}
with torch.no_grad():
test_out = self.model.get_text_features(**test_inputs)
test_tensor = self._to_tensor(test_out)
actual_dim = test_tensor.shape[-1]
if actual_dim != self.config.embedding_dim:
logger.info(
f"Model embedding dim = {actual_dim} "
f"(config said {self.config.embedding_dim}). Updating config."
)
self.config.embedding_dim = actual_dim
logger.info(f"Model loaded: {model_name} on {self.device} (dim={actual_dim})")
return
except Exception as e:
logger.warning(f"Failed to load {model_name}: {e}")
continue
raise RuntimeError(
"Could not load any CLIP model. Check internet connection and HF_TOKEN."
)
@staticmethod
def _to_tensor(output) -> torch.Tensor:
if isinstance(output, torch.Tensor):
return output
if hasattr(output, 'pooler_output') and output.pooler_output is not None:
return output.pooler_output
if hasattr(output, 'last_hidden_state'):
return output.last_hidden_state.mean(dim=1)
if hasattr(output, 'text_embeds'):
return output.text_embeds
if hasattr(output, 'image_embeds'):
return output.image_embeds
if isinstance(output, (tuple, list)) and len(output) > 0:
return output[0] if isinstance(output[0], torch.Tensor) else output[1]
raise TypeError(
f"Cannot extract tensor from model output of type {type(output)}. "
f"Available attributes: {[a for a in dir(output) if not a.startswith('_')]}"
)
@torch.no_grad()
def encode_texts(self, texts: List[str], batch_size: Optional[int] = None) -> np.ndarray:
batch_size = batch_size or min(self.config.embed_batch_size * 4, 256)
texts = [str(t) if t and str(t) != 'nan' else '' for t in texts]
all_emb = []
for i in range(0, len(texts), batch_size):
batch = texts[i:i + batch_size]
inputs = self.processor(
text=batch, return_tensors="pt",
padding=True, truncation=True, max_length=77,
)
inputs = {k: v.to(self.device) for k, v in inputs.items()}
raw = self.model.get_text_features(**inputs)
feats = self._to_tensor(raw)
feats = F.normalize(feats, p=2, dim=-1).cpu().numpy()
all_emb.append(feats)
return np.vstack(all_emb).astype(np.float32)
@torch.no_grad()
def encode_images_from_paths(
self, paths: List[Path], batch_size: Optional[int] = None,
) -> np.ndarray:
batch_size = batch_size or self.config.embed_batch_size
n = len(paths)
dim = self.config.embedding_dim
embeddings = np.zeros((n, dim), dtype=np.float32)
for start in range(0, n, batch_size):
end = min(start + batch_size, n)
batch_paths = paths[start:end]
images = []
valid_in_batch = []
for j, p in enumerate(batch_paths):
try:
img = Image.open(p).convert("RGB")
images.append(img)
valid_in_batch.append(start + j)
except Exception:
pass
if not images:
continue
try:
inputs = self.processor(images=images, return_tensors="pt", padding=True)
inputs = {k: v.to(self.device) for k, v in inputs.items()}
if self.device == "cuda":
with torch.amp.autocast("cuda"):
raw = self.model.get_image_features(**inputs)
else:
raw = self.model.get_image_features(**inputs)
feats = self._to_tensor(raw)
feats = F.normalize(feats, p=2, dim=-1).cpu().numpy()
for local_j, global_j in enumerate(valid_in_batch):
embeddings[global_j] = feats[local_j]
except Exception as e:
logger.warning(f"Batch encoding failed at {start}: {e}")
if self.device == "cuda" and start % (batch_size * 10) == 0:
torch.cuda.empty_cache()
return embeddings
@torch.no_grad()
def encode_images(self, images: List[Image.Image], batch_size: Optional[int] = None) -> np.ndarray:
batch_size = batch_size or self.config.embed_batch_size
all_emb = []
for i in range(0, len(images), batch_size):
batch = images[i:i + batch_size]
inputs = self.processor(images=batch, return_tensors="pt", padding=True)
inputs = {k: v.to(self.device) for k, v in inputs.items()}
if self.device == "cuda":
with torch.amp.autocast("cuda"):
raw = self.model.get_image_features(**inputs)
else:
raw = self.model.get_image_features(**inputs)
feats = self._to_tensor(raw)
all_emb.append(F.normalize(feats, p=2, dim=-1).cpu().numpy())
return np.vstack(all_emb).astype(np.float32)
@torch.no_grad()
def encode_query_text(self, query: str) -> np.ndarray:
prompted = [tmpl.format(query) for tmpl in self.config.prompt_templates]
embeddings = self.encode_texts(prompted)
avg = embeddings.mean(axis=0, keepdims=True)
avg = avg / (np.linalg.norm(avg, axis=-1, keepdims=True) + 1e-8)
return avg.astype(np.float32)
@torch.no_grad()
def encode_multimodal_query(
self, text: str, image: Image.Image, text_weight: float = 0.5,
) -> np.ndarray:
text_emb = self.encode_query_text(text)
img_emb = self.encode_images([image])
fused = text_weight * text_emb + (1 - text_weight) * img_emb
fused = fused / (np.linalg.norm(fused, axis=-1, keepdims=True) + 1e-8)
return fused.astype(np.float32)
|