File size: 34,059 Bytes
747451d | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 | # /*---------------------------------------------------------------------------------------------
# * Copyright (c) 2022-2023 STMicroelectronics.
# * All rights reserved.
# *
# * This software is licensed under terms that can be found in the LICENSE file in
# * the root directory of this software component.
# * If no LICENSE file comes with this software, it is provided AS-IS.
# *--------------------------------------------------------------------------------------------*/
import os
from typing import Dict, List
from munch import DefaultMunch
from omegaconf import DictConfig
from pathlib import Path
import re
import numpy as np
import requests
from hydra.core.hydra_config import HydraConfig
aspect_ratio_dict = {"fit": "ASPECT_RATIO_FIT",
"crop": "ASPECT_RATIO_CROP",
"padding": "ASPECT_RATIO_PADDING",
"full_screen": "ASPECT_RATIO_FULLSCREEN"}
color_mode_n6_dict = {"rgb": "COLOR_RGB",
"bgr": "COLOR_BGR"}
def download_file(url:str, local_path:str):
"""
Downloads a file from the given URL and saves it to the specified local path.
args:
url (str): URL of the file to download
local_path(str): Local path where the file should be saved
"""
try:
# Send a GET request to the URL
response = requests.get(url, stream=True, timeout=20) # timeout of 20 seconds
# Check if the request was successful
response.raise_for_status()
# Open the local file in write-binary mode
with open(local_path, 'wb') as file:
# Iterate over the response content in chunks
for chunk in response.iter_content(chunk_size=8192):
# Write each chunk to the local file
file.write(chunk)
print(f"[INFO] : Pretrained model file downloaded successfully and saved as :\n\t{local_path}")
except requests.exceptions.HTTPError as http_err:
print(f"[ERROR] : HTTP error occurred :\n\t{http_err}")
except Exception as err:
print(f"[ERROR] : An error occurred while downloading the pretrained model \n\t: {err}")
def check_attributes(cfg: Dict,
expected: List[str] = None,
optional: List[str] = [],
section: str = None) -> None:
'''
Checks that all the expected attributes are present in the configuration dictionary and that
there are no unknown attributes. Optional attributes may also be present.
args:
cfg (dict): The configuration dictionary.
expected (list): A list of expected (required) attributes.
optional (list, optional): A list of optional attributes. Defaults to [].
section (str, optional): The name of the config file section to check. Defaults to None.
raises:
ValueError: If an unknown or unsupported attribute is found or if a required attribute is missing.
'''
if section is not None:
message = "\nPlease check the '{}' section of your configuration file.".format(section)
else:
message = "\nPlease check your configuration file."
if cfg is not None:
if type(cfg) != DefaultMunch:
raise ValueError("Expecting an attribute. Received {}{}".format(cfg, message))
# Check that each attribute name is legal.
for attr in cfg.keys():
if (not attr in expected) and (not attr in optional):
raise ValueError("\nUnknown or unsupported attribute. Received `{}`{}".format(attr, message))
# Get the list of used attributes
used = list(cfg.keys())
else:
used = []
# Check that all the mandatory attributes are present.
for attr in expected:
if attr not in used:
raise ValueError("\nMissing `{}` attribute{}".format(attr, message))
if cfg[attr] is None:
raise ValueError("\nMissing a value for attribute `{}`{}".format(attr, message))
def collect_callback_args(name,
args=None,
message=None) -> str:
if args:
if type(args) != DefaultMunch:
raise ValueError(f"\nInvalid syntax for `{name}` callback arguments{message}")
text = "("
for k, v in args.items():
if type(v) == str and v[:7] != "lambda ":
text += f'{k}=r"{v}", '
else:
text += f'{k}={v}, '
text = text[:-2] + ")"
else:
text = "()"
return text
def get_random_seed(cfg: DictConfig = None):
'''
Returns a random seed based on the configuration file.
args:
cfg (DictConfig): The configuration object.
returns:
int or None: The random seed. If no seed is set in the configuration file, returns 0.
'''
if "global_seed" in cfg.general:
seed = cfg.general["global_seed"]
if seed == "None":
seed = None
else:
seed = int(seed)
else:
seed = 0
return seed
def replace_none_string(dico: dict) -> dict:
'''
Replaces None strings in the values of a dictionary with the Python None value.
Other values are unchanged.
args:
dico (dict): any dictionary.
'''
for k, v in dico.items():
if v == "None":
dico[k] = None
return dico
def expand_env_vars(string: str) -> str:
'''
Expands environment variables in a string if any. The syntax for variables
is ${variable_name}. An error is thrown if a variable is found in the string
but is not set.
args:
string (str): The string to expand environment variables in.
returns:
string (str): The original string with expanded variables.
'''
for match in re.findall(r'\$\{\w+\}', string):
var_name = match[2:-1]
# Get the variable value, throw an error if it is not set.
var_value = os.environ.get(var_name)
if var_value is None:
raise OSError("\nCould not find an environment variable named `{}`\n"
"Please check your configuration file.".format(var_name))
match = "\\" + match
string = re.sub(match, var_value, string, count=1)
return string
def postprocess_config_dict(config: DictConfig, replace_none_string=False) -> None:
'''
The YAML loader outputs some attribute values as strings although they
are different Python types. This function walks the config dictionary
tree and restores some of these types, including booleans, integers,
floats and tuples.
For example:
- "True" is converted to boolean True
- "1./255" is converted to a float (equal to 0.00392156)
- "(128, 128, 3)" is converted to a tuple.
The function also replaces environment variables that appear in strings
with their values.
args:
config (DictConfig): dictionary containing the entire configuration file.
'''
for k in config.keys():
v = config[k]
if type(v) == dict:
postprocess_config_dict(v, replace_none_string=replace_none_string)
elif type(v) == str:
if replace_none_string and v.lower() == "none":
config[k] = None
continue
# Expand environment variables if any
v_exp = expand_env_vars(v)
if v_exp != v:
config[k] = v_exp
v = v_exp
if v[:7] == "lambda ":
# The value is a lambda function. Remove the \n characters
# and multiple blanks that get inserted by the YAML loader
# if the function is written on several lines.
v = re.sub("\n", "", v)
config[k] = re.sub(" +", " ", v)
else:
try:
v_eval = eval(v)
except:
v_eval = v
if isinstance(v_eval, (bool, int, float, tuple)):
config[k] = v_eval
def check_config_attributes(cfg: DictConfig,
specs: Dict = None,
section: str = None) -> None:
'''
This function checks that the attributes used in a given section
of the configuration file comply with specified requirements.
args:
cfg (DictConfig): dictionary containing the configuration file section to check
specs (Dict): dictionary specifying the requirements for attribute usage in the section
section (str): name of the section
'''
specs = DefaultMunch.fromDict(specs)
if section == "top_level":
message = f"\nPlease check the top-level of your configuration file."
else:
message = f"\nPlease check the '{section}' section of your configuration file."
if specs.legal:
# Check that all the used attribute names are known
for attr in cfg.keys():
if attr not in specs.legal:
raise ValueError(f"\nUnknown attribute `{attr}`{message}")
if specs.all:
# Check that all the specified attributes are present and have a value
for attr in specs.all:
if attr not in cfg:
if section == "top_level":
raise ValueError(f"\nMissing `{attr}` section{message}")
else:
raise ValueError(f"\nMissing `{attr}` attribute{message}")
if cfg[attr] is None:
if section == "top_level":
raise ValueError(f"\nMissing body of `{attr}` section{message}")
else:
raise ValueError(f"\nExpecting a value for `{attr}` attribute{message}")
if specs.one_or_more:
# Check that at least one of the specified attributes is present and has a value
count = 0
for attr in specs.one_or_more:
if attr in cfg and cfg[attr] is not None:
count += 1
if count == 0:
raise ValueError(f"\nMissing one or more attributes from {specs.one_or_more}{message}")
def check_model_file_extension(ml_path, mode, mode_groups, field_name):
"""
Validates the file extension and existence of a model file path according to the current operation mode.
Args:
ml_path (str): Path to the model file to check.
mode (str): The current operation mode (e.g., 'training', 'quantization', etc.).
mode_groups (Any): An object with attributes for each mode group, each being a list of mode names.
field_name (str): The name of the config field being checked (for error messages).
Raises:
ValueError: If the file extension is not allowed for the current mode, or if the path is not provided.
FileNotFoundError: If the file does not exist at the given path.
"""
m1 = f"\nExpecting `{field_name}` attribute to be set to a path to "
m2 = "\nPlease check the 'model' section of your configuration file."
if not ml_path:
raise ValueError(m1 + "a valid file path" + m2)
file_extension = Path(ml_path).suffix.lower()
if mode in mode_groups.training:
allowed = [".h5", ".keras"]
if file_extension not in allowed:
raise ValueError(m1 + ", ".join(allowed) + m2)
elif mode in mode_groups.quantization:
allowed = [".h5", ".keras", ".onnx"]
if file_extension not in allowed:
raise ValueError(m1 + ", ".join(allowed) + m2)
elif mode in ("evaluation", "prediction"):
allowed = [".h5", ".keras", ".tflite", ".onnx"]
if file_extension not in allowed:
raise ValueError(m1 + ", ".join(allowed) + m2)
elif mode in ("benchmarking", "deployment"):
allowed = [".h5", ".keras", ".tflite", ".onnx"]
if file_extension not in allowed:
raise ValueError(m1 + ", ".join(allowed) + m2)
if not os.path.isfile(ml_path):
raise FileNotFoundError(
f"\nUnable to find file {ml_path}\nPlease check the '{field_name}' attribute in your configuration file"
)
def parse_tools_section(cfg: DictConfig,
operation_mode: str,
hardware_type: str ="MCU") -> None:
'''
parses the 'tools' section of the configuration file
args:
cfg (DictConfig): 'tools' section of configuration file
operation_mode (str): service or operation mode used
hardware_type (str): type of hardware targetted
'''
# if cfg is not None:
required = []
if hardware_type == "MCU" and not operation_mode == "evaluation" and not operation_mode == "prediction":
required += ["path_to_cubeIDE",]
if cfg.stedgeai:
legal = ["stedgeai", "path_to_cubeIDE"]
check_config_attributes(cfg, specs={"legal": legal, "all": required}, section="tools")
# stedgeai usage
legal = ["optimization", "on_cloud", "path_to_stedgeai"]
check_config_attributes(cfg.stedgeai,
specs={"legal": legal, "all": []}, section="tools.stedgeai")
if not cfg.stedgeai.on_cloud:
if not os.path.isfile(cfg.stedgeai.path_to_stedgeai):
print(cfg.stedgeai.path_to_stedgeai)
raise ValueError("Path for `stedgeai.exe` does not exist.\n"
"Please check the cfg.tools.stedgeai section!")
# Patch to support stedgeai with legacy naming stm32ai : reconstruct stm32ai dictionnary
# from stedgeai one
cfg["stm32ai"] = cfg.stedgeai
cfg.stm32ai["optimization"] = cfg.stedgeai.optimization if cfg.stedgeai.optimization else "balanced"
cfg.stm32ai["on_cloud"] = cfg.stedgeai.on_cloud # if cfg.stedgeai.on_cloud else True
cfg.stm32ai["path_to_stm32ai"] = cfg.stedgeai.path_to_stedgeai if cfg.stedgeai.path_to_stedgeai else None
cfg.stm32ai["version"] = Path(cfg.stedgeai.path_to_stedgeai).parts[-4]
cfg.stedgeai["version"] = Path(cfg.stedgeai.path_to_stedgeai).parts[-4]
# Path to cubeIDE only needed for MCU in deployment service
if hardware_type == "MCU":
if operation_mode == "deployment" and not os.path.isfile(cfg.path_to_cubeIDE):
raise ValueError("Path for `path_to_cubeIDE` does not exist.\n"
"Please check the cfg.tools section!")
def parse_benchmarking_section(cfg: DictConfig) -> None:
'''
parses the 'benchmarking' section of the configuration file
args:
cfg (DictConfig): 'benchmarking' section of the configuration file
'''
legal = ["board"]
check_config_attributes(cfg, specs={"legal": legal, "all": legal}, section="benchmarking")
def parse_quantization_section(cfg: DictConfig,
legal: List) -> None:
'''
parses the dictionary containing entire configuration file
args:
cfg (DictConfig): 'quantization' section of the configuration file
legal (List): UC specific usable attributes
'''
required = [x for x in legal if x not in ["export_dir", "granularity", "optimize", "target_opset", "operating_mode",
"onnx_quant_parameters", "op_types_to_quantize", "onnx_extra_options",
"iterative_quant_parameters"]]
check_config_attributes(cfg, specs={"legal": legal, "all": required}, section="quantization")
# Set default values of missing optional arguments
if not cfg.export_dir:
cfg.export_dir = "quantized_models"
if not cfg.granularity:
cfg.granularity = "per_channel"
cfg.optimize = cfg.optimize if cfg.optimize is not None else False
cfg.target_opset = cfg.target_opset if cfg.target_opset is not None else 17
cfg.operating_mode = cfg.operating_mode if cfg.operating_mode else 'default'
# Check the quantizer name
if cfg.quantizer.lower() not in ["tflite_converter", "onnx_quantizer"]:
raise ValueError(f"\nUnknown or unsupported quantizer. Received `{cfg.quantizer}`\n"
"Supported quantizers are : TFlite_converter or Onnx_quantizer\n"
"Please check the 'quantization.quantizer' attribute in your configuration file.")
# Check the granularity value
if cfg.granularity not in ['per_channel', 'per_tensor']:
raise ValueError(f"\nUnknown or unsupported granularity value. Received `{cfg.granularity}`\n"
"Supported granularity: 'per_channel, or 'per_tensor'\n"
"Please check the 'quantization.granularity' attribute in your configuration file.")
# Check optimization
if cfg.optimize not in [True, False]:
raise ValueError(f"\nUnknown or unsupported optimize value. Received `{cfg.optimize}`\n"
"Supported optimize parameters: 'True', or 'False'\n"
"Please check the 'quantization.optimize' attribute in your configuration file.")
if not isinstance(cfg.target_opset, int):
raise ValueError(f"\nUnknown or unsupported target_opset value. Received `{cfg.optimize}`\n"
"Supported target_opset parameters: 'int' up to latest onnx_opset\n"
"Please check the 'quantization.target_opset' attribute in your configuration file.")
# Check the quantizer type
if cfg.quantization_type.lower() not in ["ptq"]:
raise ValueError(f"\nUnknown or unsupported quantization type. Received `{cfg.quantization_type}`\n"
"Supported type: PTQ\n"
"Please check the 'quantization.quantization_type' attribute in your configuration file.")
def parse_evaluation_section(cfg: DictConfig,
legal: List) -> None:
'''
parses the dictionary containing entire configuration file
args:
cfg (DictConfig): 'evaluation' section of the configuration file
legal (List): UC specific usable attributes
'''
required = []
check_config_attributes(cfg, specs={"legal": legal, "all": required}, section="evaluation")
# Set default values of missing optional arguments
if not cfg.gen_npy_input:
cfg.gen_npy_input = False
if not cfg.gen_npy_output:
cfg.gen_npy_output = False
if not cfg.profile:
cfg.profile = "profile_O3"
if not cfg.input_type:
cfg.input_type = "uint8"
if not cfg.output_type:
cfg.output_type = "int8"
if not cfg.input_chpos:
cfg.input_chpos = "chlast"
if not cfg.output_chpos:
cfg.output_chpos = "chlast"
if not cfg.target:
cfg.target = "host"
def parse_top_level(cfg: DictConfig,
mode_groups: DictConfig = None,
mode_choices: List = None,
legal: List = None) -> None:
'''
parses the dictionary containing entire configuration file
args:
cfg (DictConfig): configuration dictionary containing entire configuration file
mode_groups (DictConfig): configuration about a given mode
mode_choices (List): currently supported modes
legal (List): UC specific usable attributes
'''
# Check that operation_mode is present and has a value
message = "\nPlease check the top-level of your configuration file."
if "operation_mode" not in cfg:
raise ValueError("\nMissing `operation_mode` attribute\n"
f"Supported modes: {mode_choices}{message}")
if cfg.operation_mode is None:
raise ValueError("\nExpecting a value for `operation_mode` attribute\n"
f"Supported modes: {mode_choices}{message}")
# [KH]: to be added when all use cases have model section]
# # Check that the model section is present and has a value
# if "model" not in cfg:
# raise ValueError("\nMissing `model` section at the top level of your configuration file.\n"
# f"Please check your configuration file.{message}")
# if cfg.model is None:
# raise ValueError("\nExpecting a value for `model` section at the top level of your configuration file.\n"
# f"Please check your configuration file.{message}")
# Check that the value of operation_mode is valid
mode = cfg.operation_mode
if mode not in mode_choices:
raise ValueError(f"\nUnknown value for `operation_mode` attribute. Received {mode}\n"
f"Supported modes: {mode_choices}{message}")
# Attributes usable at the top level
required = ["mlflow"] # [KH]: should include model later] #, "model"]
# if mode not in mode_groups.training:
# # We need the 'general' section to provide model_path.
# required += ["general",]
if mode != "benchmarking":
# Need the preprocessing & feature extraction sections (when available)
required += ["preprocessing"]
if "feature_extraction" in legal:
required += ["feature_extraction"]
if mode not in ("prediction", "quantization", "benchmarking", "deployment", "chain_qb", "chain_qd"):
required += ["dataset",]
if mode in mode_groups.training:
required += ["training",]
if mode in mode_groups.quantization:
required += ["quantization",]
# if mode == "prediction":
# required += ["prediction",]
if mode in mode_groups.benchmarking:
required += ["benchmarking", "tools"]
if mode in mode_groups.deployment:
required += ["deployment", "tools"]
if mode in mode_groups.compression:
required += ["compression", "training"] # Needed as fine tuning is part of the compression feature
check_config_attributes(cfg, specs={"legal": legal, "all": required}, section="top_level")
def parse_general_section(cfg: DictConfig,
mode: str = None,
mode_groups: str = None,
legal: List = None,
required: List = None,
output_dir: str = '') -> None:
'''
parses the general section of configuration file.
args:
cfg (DictConfig): configuration dictionary
mode (str): operation mode
mode_groups (str): operation mode group
legal (List): UC specific usable attributes
required (List): UC specific required attributes
output_dir (str): output directory for the current run
'''
# # Usage of the model_path attribute in training modes
# # is checked when parsing the 'training' section.
# required.append("model_path") if not mode_groups.training else []
check_config_attributes(cfg, specs={"legal": legal, "all": required}, section="general")
# Set default values of missing optional attributes
if not cfg.project_name:
cfg.project_name = "<unnamed>"
if not cfg.logs_dir:
cfg.logs_dir = "logs"
if not cfg.saved_models_dir:
cfg.saved_models_dir = "saved_models"
cfg.deterministic_ops = cfg.deterministic_ops if cfg.deterministic_ops is not None else False
cfg.display_figures = cfg.display_figures if cfg.display_figures is not None else True
if not cfg.global_seed or cfg.global_seed == 'None':
cfg.global_seed = 123
if not cfg.num_threads_tflite:
cfg.num_threads_tflite = 1
def parse_random_periodic_resizing(cfg, output_stride):
message = "Please check the 'random_periodic_resizing' section of your configuration file."
if "period" not in cfg:
raise ValueError(f"\nMissing `period` argument\n{message}")
if "image_sizes" not in cfg:
raise ValueError(f"\nMissing `image_sizes` argument\n{message}")
# Image sizes can be given using tuples, arrays or a mix
# of tuples and arrays. We convert all sizes to tuples.
sizes_str = '['
for size in cfg.image_sizes:
if isinstance(size, (list, tuple)):
sizes_str += '('
for x in size:
sizes_str += str(x) + ','
sizes_str = sizes_str[:-1] + '),'
else:
sizes_str += str(size) + ','
sizes_str = sizes_str[:-1] + ']'
sizes_message = "\nInvalid syntax for `image_sizes` argument\n"
try:
x = eval(sizes_str)
random_sizes = np.array(x, dtype=np.int32)
except:
raise ValueError(sizes_message + message)
if np.shape(random_sizes)[1] != 2:
raise ValueError(sizes_message + message)
# Check that the image sizes are compatible with the network stride.
for size in random_sizes:
if np.shape(output_stride)==():
output_strides = [output_stride]
else:
output_strides = output_stride
for os in output_strides:
if (size[0] % os != 0) or (size[1] % os != 0):
raise ValueError(
f"Image sizes must be multiples of the network stride.\n"
f"Network stride: {os}\n"
f"Invalid image size: {size}\n"
f"{message}")
return random_sizes.tolist()
def parse_compression_section(cfg: DictConfig,
legal: List) -> None:
'''
parses the dictionary containing entire configuration file
args:
cfg (DictConfig): 'compression' section of the configuration file
legal (List): UC specific usable attributes
'''
required = [x for x in legal if x not in ["factor", "strong_optimization"]]
check_config_attributes(cfg, specs={"legal": legal, "all": required}, section="compression")
# Set default values of missing optional arguments
cfg.factor = cfg.factor if cfg.factor else 0.5
cfg.strong_optimization = cfg.strong_optimization if cfg.strong_optimization else False
# Check the compression factor type
if not isinstance(cfg.factor, float):
raise ValueError(f"\nUnknown or unsupported factor value. Received `{cfg.factor}`\n"
"Supported factor parameters: 'float'\n"
"Please check the 'compression.factor' attribute in your configuration file.")
# Check optimization
if cfg.strong_optimization not in [True, False]:
raise ValueError(f"\nUnknown or unsupported strong_optimization value. Received `{cfg.strong_optimization}`\n"
"Supported optimize parameters: 'True', or 'False'\n"
"Please check the 'compression.strong_optimization' attribute in your configuration file.")
def parse_training_section(cfg: DictConfig,
legal: List = None) -> None:
'''
parses the training section of configuration file.
args:
cfg (DictConfig): 'training' section of the configuration file
model_path_used (bool): a flag to tell if the 'model.model_path' parameter is provided
model_type_used (bool): a flag to tell if the 'model.model_type' parameter is provided
legal (List): usable attributes
'''
required = ["batch_size", "epochs", "optimizer"]
check_config_attributes(cfg, specs={"legal": legal, "all": required}, section="training")
# The optimizer may be written on one line. For example: "optimizer: Adam"
# In this case, we got a string instead of a dictionary.
if type(cfg.optimizer) == str:
cfg.optimizer = DefaultMunch.fromDict({cfg.optimizer: None})
def parse_model_section(cfg: DictConfig, mode: str, mode_groups, legal: list, required: list = None) -> None:
"""
Checks and parses the root-level 'model' section of the config file.
Args:
cfg (DictConfig): The model configuration parameters.
mode (str): The current operation mode.
mode_groups: The mode groups object.
legal (list): List of legal model attributes for this use case.
required (list, optional): List of required model attributes. If None, no required fields are enforced.
Returns:
None
"""
req = [] if required is None else list(required)
if cfg.model_name:
req.append("input_shape")
# Mutually exclusive model sources
model_sources = ["model_name", "model_path"]
set_sources = [name for name in model_sources if getattr(cfg, name, None)]
if len(set_sources) == 0:
raise ValueError(
"\nExpecting one of the following model source attributes to be set: "
f"{', '.join(model_sources)}\nPlease check your configuration file." )
check_config_attributes(cfg, specs={"legal": legal, "all": req}, section="model")
if cfg.model_path and cfg.model_path[:4].lower() == "http":
print('[INFO] : A URL found for model.model_path variable!')
url = cfg.model_path
output_dir = HydraConfig.get().runtime.output_dir
model_dir = os.path.join(output_dir, os.path.splitext(os.path.basename(cfg.model_path))[0])
os.makedirs(model_dir, exist_ok=True)
local_path = os.path.join(model_dir, url.split('/')[-1])
download_file(url, local_path)
cfg.model_path = local_path
# Check model_path
if cfg.model_path:
file_extension = Path(cfg.model_path).suffix.lower()
if file_extension in [".h5", ".keras", ".tflite"]:
cfg.framework = "tf"
check_model_file_extension(cfg.model_path, mode, mode_groups, "model_path")
elif file_extension in [".pt", ".pth"]:
cfg.framework = "torch"
elif file_extension in [".onnx"] and not cfg.framework:
cfg.framework = "tf"
check_model_file_extension(cfg.model_path, mode, mode_groups, "model_path")
else:
# Get end of model name to set the framework to be used"
if cfg.model_name[-3:]=='_pt':
cfg.framework = "torch"
else:
cfg.framework = "tf"
if cfg.framework == "tf":
if len(set_sources) > 1:
raise ValueError(
"\nThe following model source attributes are mutually exclusive and more than one is set: "
f"{', '.join(set_sources)}\nPlease check your configuration file.")
def parse_prediction_section(cfg: DictConfig) -> None:
'''
parses the prediction section of configuration file.
args:
cfg (DictConfig): 'prediction' section of the configuration file
'''
legal = ["seed","target", "reid_distance_metric",
"profile", "input_type", "output_type", "input_chpos", "output_chpos"]
required = []
check_config_attributes(cfg, specs={"legal": legal, "all": required}, section="prediction")
# Set default values of missing optional arguments
if not cfg.profile:
cfg.profile = "profile_O3"
if not cfg.input_type:
cfg.input_type = "uint8"
if not cfg.output_type:
cfg.output_type = "int8"
if not cfg.input_chpos:
cfg.input_chpos = "chlast"
if not cfg.output_chpos:
cfg.output_chpos = "chlast"
if not cfg.target:
cfg.target = "host"
def parse_deployment_section(cfg: DictConfig,
legal: List = None,
legal_hw: List = None) -> None:
'''
parses the training section of configuration file.
args:
cfg (DictConfig): 'deployment' section of the configuration file
legal (List): usable attributes
legal_hw (List): usable attributes for the HW setup part
'''
required = [x for x in legal]
check_config_attributes(cfg, specs={"legal": legal, "all": required}, section="deployment")
required = [x for x in legal_hw if x != 'stlink_serial_number']
check_config_attributes(cfg.hardware_setup, specs={"legal": legal_hw, "all": required},
section="deployment.hardware_setup")
def parse_mlflow_section(cfg: DictConfig) -> None:
'''
parses the mlflow section of configuration
args:
cfg (DictConfig): 'mlflow' section of the configuration
'''
legal = ["uri"]
check_config_attributes(cfg, specs={"legal": legal, "all": legal}, section="mlflow")
def check_hardware_type(cfg: DictConfig,
mode_groups: DictConfig = None) -> None:
'''
parses the mlflow section of configuration
args:
cfg (DictConfig): dictionary containing the configuration file section to check
mode_groups (DictConfig): configuration about a given mode
'''
# By default MCU is selected
cfg["hardware_type"] = "MCU"
# Check if a MPU target is specified in the configuration
if cfg.operation_mode in mode_groups.benchmarking:
if cfg.benchmarking.board is not None:
if "STM32MP" in cfg.benchmarking.board:
cfg.hardware_type = "MPU"
elif cfg.operation_mode in mode_groups.deployment:
if cfg.deployment.hardware_setup.serie is not None:
if "STM32MP" in cfg.deployment.hardware_setup.serie:
cfg.hardware_type = "MPU"
def get_class_names_from_file(cfg: DictConfig) -> List[str]:
if cfg.classes_file_path :
with open(cfg.classes_file_path, 'r') as file:
class_names = [line.strip() for line in file]
return class_names
from omegaconf import DictConfig, OmegaConf
def flatten_config(cfg, preserve_keys=("class_map",)):
# Ensure cfg is a plain dict first
if isinstance(cfg, DictConfig):
cfg = OmegaConf.to_container(cfg, resolve=False)
elif "DefaultMunch" in str(type(cfg)):
cfg = cfg.toDict()
else:
cfg = dict(cfg)
preserve_keys = set(preserve_keys)
flat_config = {}
def _flatten(d):
for k, v in d.items():
if isinstance(v, dict):
if k in preserve_keys:
flat_config[k] = v
else:
_flatten(v)
else:
flat_config[k] = v
_flatten(cfg)
return flat_config |