File size: 13,858 Bytes
18faf97 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 |
# ----------------------------------------------------------------------
# IMPORTS
# ----------------------------------------------------------------------
import torch
import logging
import time
import spaces
import sys
import traceback
from PIL import Image
from typing import List, Optional, Any
from collections import defaultdict
from src.utils import LOG_LEVEL_MAP, EMOJI_MAP
# ----------------------------------------------------------------------
# RT-DETR CONSTANTS
# ----------------------------------------------------------------------
RTDETR_CONF = 0.4
RTDETR_ARTIFACT_CONF = 0.35
# ----------------------------------------------------------------------
# MODEL LABEL CONFIGURATION
# ----------------------------------------------------------------------
MODEL_LABEL_CONFIG = {
"rtdetr_model": {
"person_list": {
"person": ["person"]
},
"product_type_list": {},
"head_list": {},
"shoes_list": {},
"clothing_features_list": {
"collar": ["tie"]
},
"artifacts_list": {
"bag": ["backpack", "handbag", "suitcase"],
"cup": ["bottle", "wine glass", "cup"],
"umbrella": ["umbrella"],
"book": ["book"],
"phone": ["cell phone"],
"camera": [],
"other": ["fork", "knife", "spoon", "bowl", "frisbee", "sports ball", "kite",
"baseball bat", "baseball glove", "skateboard", "surfboard", "tennis racket",
"chair", "couch", "potted plant", "bed", "dining table", "toilet", "tv",
"laptop", "mouse", "remote", "keyboard", "microwave", "oven", "toaster",
"sink", "refrigerator", "clock", "vase", "scissors", "teddy bear",
"hair drier", "toothbrush"]
}
}
}
# ----------------------------------------------------------------------
# RT-DETR HELPER FUNCTIONS
# ----------------------------------------------------------------------
def get_rtdetr_clothing_labels():
clothing_labels = set()
rtdetr_config = MODEL_LABEL_CONFIG.get("rtdetr_model", {})
for keyword, labels in rtdetr_config.get("person_list", {}).items():
clothing_labels.update(labels)
for keyword, labels in rtdetr_config.get("product_type_list", {}).items():
clothing_labels.update(labels)
clothing_labels.update(["coat", "dress", "jacket", "shirt", "skirt", "pants", "shorts"])
return clothing_labels
def get_rtdetr_person_and_product_labels():
labels = set()
rtdetr_config = MODEL_LABEL_CONFIG.get("rtdetr_model", {})
for keyword, label_list in rtdetr_config.get("person_list", {}).items():
labels.update(label_list)
for keyword, label_list in rtdetr_config.get("product_type_list", {}).items():
labels.update(label_list)
labels.update(["person", "coat", "dress", "jacket", "shirt", "skirt", "pants", "shorts"])
return labels
def get_rtdetr_artifact_labels():
artifact_labels = set()
rtdetr_config = MODEL_LABEL_CONFIG.get("rtdetr_model", {})
for keyword, labels in rtdetr_config.get("artifacts_list", {}).items():
if keyword != "other":
artifact_labels.update(labels)
return artifact_labels
def get_label_name_from_model(model, label_id):
if hasattr(model, 'config') and hasattr(model.config, 'id2label'):
return model.config.id2label.get(label_id, f"unknown_{label_id}").lower()
if hasattr(model, 'model_labels') and isinstance(model.model_labels, dict):
return model.model_labels.get(label_id, f"unknown_{label_id}").lower()
return f"unknown_{label_id}"
def map_label_to_keyword(label_name: str, valid_kws: List[str], model_name: str) -> Optional[str]:
ln = label_name.strip().lower()
model_config = MODEL_LABEL_CONFIG.get(model_name, {})
for list_type in ["person_list", "product_type_list", "head_list",
"shoes_list", "clothing_features_list", "artifacts_list"]:
category_config = model_config.get(list_type, {})
for keyword, labels in category_config.items():
if keyword in valid_kws:
for label in labels:
if ln == label.lower() or ln in label.lower():
return keyword
return None
def process_rtdetr_results(results, model, label_set, threshold, fallback_box=None):
try:
if isinstance(results, list):
if len(results) > 0:
result = results[0]
else:
return None, 0.0, None
else:
result = results
found_box = None
found_score = 0.0
found_label = None
for score, label, box in zip(result["scores"], result["labels"], result["boxes"]):
score_val = score.item()
if score_val < threshold:
continue
label_id = label.item()
label_name = get_label_name_from_model(model, label_id)
if label_name in label_set:
x1, y1, x2, y2 = [int(val) for val in box.tolist()]
found_box = [x1, y1, x2, y2]
found_score = score_val
found_label = label_name
break
return found_box, found_score, found_label
except Exception as e:
logging.log(LOG_LEVEL_MAP["WARNING"], f"{EMOJI_MAP['WARNING']} Error processing RTDETR results: {e}")
return fallback_box, 0.0, None
# ----------------------------------------------------------------------
# RT-DETR DETECTION FUNCTIONS
# ----------------------------------------------------------------------
def detect_rtdetr_in_roi(roi_rgb, RTDETR_PROCESSOR, RTDETR_MODEL, DEVICE, log_item):
boxes = []
labels = []
scores = []
raw_labels = []
try:
rtdetr_inputs = RTDETR_PROCESSOR(images=roi_rgb, return_tensors="pt")
rtdetr_inputs = {k: v.to(DEVICE) for k, v in rtdetr_inputs.items()}
with torch.no_grad():
rtdetr_outputs = RTDETR_MODEL(**rtdetr_inputs)
rtdetr_results = RTDETR_PROCESSOR.post_process_object_detection(
rtdetr_outputs,
target_sizes=torch.tensor([[roi_rgb.height, roi_rgb.width]]).to(DEVICE),
threshold=RTDETR_CONF
)
if isinstance(rtdetr_results, list) and len(rtdetr_results) > 0:
result = rtdetr_results[0]
else:
result = rtdetr_results
for score, label, box in zip(result["scores"], result["labels"], result["boxes"]):
label_id = label.item()
score_val = score.item()
x1, y1, x2, y2 = [int(val) for val in box.tolist()]
label_name = get_label_name_from_model(RTDETR_MODEL, label_id)
boxes.append([x1, y1, x2, y2])
labels.append(label_id)
scores.append(score_val)
raw_labels.append(label_name)
logging.log(LOG_LEVEL_MAP["INFO"], f"rtdetr_model: {EMOJI_MAP['INFO']} RT-DETR detected: {label_name} at score {score_val:.3f}")
except Exception as e:
error_msg = f"RTDETR detection error: {str(e)}"
error_trace = traceback.format_exc()
logging.log(LOG_LEVEL_MAP["WARNING"], f"{EMOJI_MAP['WARNING']} {error_msg}")
logging.error(f"Traceback:\n{error_trace}")
log_item["warnings"] = log_item.get("warnings", []) + [error_msg]
log_item["traceback"] = error_trace
if "CUDA must not be initialized" in str(e):
logging.critical("CUDA initialization error in Spaces Zero GPU environment")
sys.exit(1)
return boxes, labels, scores, raw_labels
def detect_rtdetr_artifacts_in_roi(roi_rgb, keywords, RTDETR_PROCESSOR, RTDETR_MODEL, DEVICE, log_item):
boxes = []
labels = []
scores = []
raw_labels = []
try:
rtdetr_inputs = RTDETR_PROCESSOR(images=roi_rgb, return_tensors="pt")
rtdetr_inputs = {k: v.to(DEVICE) for k, v in rtdetr_inputs.items()}
with torch.no_grad():
rtdetr_outputs = RTDETR_MODEL(**rtdetr_inputs)
rtdetr_results = RTDETR_PROCESSOR.post_process_object_detection(
rtdetr_outputs,
target_sizes=torch.tensor([[roi_rgb.height, roi_rgb.width]]).to(DEVICE),
threshold=RTDETR_ARTIFACT_CONF
)
rtdetr_artifact_labels = get_rtdetr_artifact_labels()
if isinstance(rtdetr_results, list) and len(rtdetr_results) > 0:
result = rtdetr_results[0]
else:
result = rtdetr_results
for score, label, box in zip(result["scores"], result["labels"], result["boxes"]):
label_id = label.item()
score_val = score.item()
if score_val < RTDETR_ARTIFACT_CONF:
continue
label_name = get_label_name_from_model(RTDETR_MODEL, label_id)
if label_name in rtdetr_artifact_labels:
x1, y1, x2, y2 = [int(val) for val in box.tolist()]
artifact_keyword = map_label_to_keyword(label_name, keywords, "rtdetr_model")
if not artifact_keyword:
continue
boxes.append([x1, y1, x2, y2])
labels.append(label_id)
scores.append(score_val)
raw_labels.append(label_name)
logging.log(LOG_LEVEL_MAP["INFO"], f"rtdetr_model: {EMOJI_MAP['INFO']} Artifact detected: {label_name} at score {score_val:.3f}")
except Exception as e:
error_msg = f"RTDETR artifact detection error: {str(e)}"
error_trace = traceback.format_exc()
logging.log(LOG_LEVEL_MAP["WARNING"], f"{EMOJI_MAP['WARNING']} {error_msg}")
logging.error(f"Traceback:\n{error_trace}")
log_item["warnings"] = log_item.get("warnings", []) + [error_msg]
log_item["traceback"] = error_trace
if "CUDA must not be initialized" in str(e):
logging.critical("CUDA initialization error in Spaces Zero GPU environment")
sys.exit(1)
return boxes, labels, scores, raw_labels
def update_fallback_detection(ctx, pi_rgba, fallback_box, RTDETR_PROCESSOR, RTDETR_MODEL, DEVICE, RTDETR_CONF, final_boxes, final_labels, final_scores, final_kws, final_raws, final_mods, dd_log):
try:
if not (fallback_box and isinstance(fallback_box, list) and len(fallback_box) == 4):
return final_boxes, final_labels, final_scores, final_kws, final_raws, final_mods, dd_log
sub_ = pi_rgba.crop((
fallback_box[0],
fallback_box[1],
fallback_box[2],
fallback_box[3]
))
sub_ = sub_.convert("RGB")
subW = sub_.width
subH = sub_.height
rtdetr_inputs = RTDETR_PROCESSOR(images=sub_, return_tensors="pt").to(DEVICE)
with torch.no_grad():
rtdetr_outputs = RTDETR_MODEL(**rtdetr_inputs)
rtdetr_results = RTDETR_PROCESSOR.post_process_object_detection(
rtdetr_outputs,
target_sizes=torch.tensor([[subH, subW]]).to(DEVICE),
threshold=RTDETR_CONF
)
rtdetr_clothing_labels = get_rtdetr_clothing_labels()
found_fb_box, found_fb_score, _ = process_rtdetr_results(
rtdetr_results, RTDETR_MODEL, rtdetr_clothing_labels, RTDETR_CONF
)
if found_fb_box:
fx1 = fallback_box[0] + found_fb_box[0]
fy1 = fallback_box[1] + found_fb_box[1]
fx2 = fallback_box[0] + found_fb_box[2]
fy2 = fallback_box[1] + found_fb_box[3]
found_fb_box = [fx1, fy1, fx2, fy2]
final_boxes.append(found_fb_box)
final_labels.append(90001)
final_scores.append(round(found_fb_score, 2))
final_kws.append(ctx.product_type)
final_raws.append("fallback_label")
final_mods.append("rtdetr_model")
dd_log[ctx.product_type].append({
"box": found_fb_box,
"score": found_fb_score,
"raw_label": "fallback_label",
"model": "rtdetr_model"
})
else:
final_boxes.append(fallback_box)
final_labels.append(90000)
final_scores.append(0.0)
final_kws.append(ctx.product_type)
final_raws.append("fallback_label")
final_mods.append("fallback")
dd_log[ctx.product_type].append({
"box": fallback_box,
"score": 0.0,
"raw_label": "fallback_label",
"model": "fallback"
})
return final_boxes, final_labels, final_scores, final_kws, final_raws, final_mods, dd_log
except Exception as e:
logging.log(LOG_LEVEL_MAP["WARNING"], f"{EMOJI_MAP['WARNING']} Fallback detection error: {e}")
final_boxes.append(fallback_box)
final_labels.append(90000)
final_scores.append(0.0)
final_kws.append(ctx.product_type)
final_raws.append("fallback_label")
final_mods.append("fallback_error")
dd_log[ctx.product_type].append({
"box": fallback_box,
"score": 0.0,
"raw_label": "fallback_label",
"model": "fallback_error"
})
return final_boxes, final_labels, final_scores, final_kws, final_raws, final_mods, dd_log
|