def run_models_for_device(flags, args, dev):
conf = config_parser.parse(flags.config)
for name, model_conf in conf["models"].items():
if not flags.model_name or name == flags.model_name:
MaceLogger.info("Run model %s" % name)
model_conf = config_parser.normalize_model_config(model_conf)
run_model_for_device(flags, args, dev, name, model_conf)
def main() -> None:
config = config_parser.parse('config.json')
db_worker = DBWorker(config)
redis_client = redis.Redis(password=config['redis_pass'])
try:
redis_client.get('1')
except redis.exceptions.ConnectionError:
print('Ошибка подключения к Redis.')
exit()
args = parse_args()
if args.s: # запуск
start(config, db_worker, redis_client)
parser.add_argument("--backend",
type=str,
default="tensorflow",
help="onnx backend framework")
parser.add_argument("--build",
action="store_true",
help="if build before run")
parser.add_argument('--output',
type=str,
default="build",
help="output dir")
parser.add_argument('--vlog_level',
type=int,
default="0",
help="vlog level")
return parser.parse_known_args()
if __name__ == "__main__":
flags, args = parse_args()
conf = config_parser.parse(flags.config)
if flags.build or flags.validate:
micro_conf = config_parser.normalize_model_config(
conf[ModelKeys.models][flags.model_name])
build_engine(flags.model_name, micro_conf[ModelKeys.data_type])
if flags.validate and flags.layers != "-1":
run_layers_validate(flags, args, conf)
else:
run_model(flags, args, conf)
def convert_func(flags):
configs = config_parser.parse(flags.config)
print(configs)
library_name = configs[YAMLKeyword.library_name]
if not os.path.exists(BUILD_OUTPUT_DIR):
os.makedirs(BUILD_OUTPUT_DIR)
elif os.path.exists(os.path.join(BUILD_OUTPUT_DIR, library_name)):
sh.rm("-rf", os.path.join(BUILD_OUTPUT_DIR, library_name))
os.makedirs(os.path.join(BUILD_OUTPUT_DIR, library_name))
if not os.path.exists(BUILD_DOWNLOADS_DIR):
os.makedirs(BUILD_DOWNLOADS_DIR)
model_output_dir = \
'%s/%s/%s' % (BUILD_OUTPUT_DIR, library_name, MODEL_OUTPUT_DIR_NAME)
model_header_dir = \
'%s/%s/%s' % (BUILD_OUTPUT_DIR, library_name, MODEL_HEADER_DIR_PATH)
# clear output dir
if os.path.exists(model_output_dir):
sh.rm("-rf", model_output_dir)
os.makedirs(model_output_dir)
if os.path.exists(model_header_dir):
sh.rm("-rf", model_header_dir)
if os.path.exists(MODEL_CODEGEN_DIR):
sh.rm("-rf", MODEL_CODEGEN_DIR)
if os.path.exists(ENGINE_CODEGEN_DIR):
sh.rm("-rf", ENGINE_CODEGEN_DIR)
if flags.quantize_stat:
configs[YAMLKeyword.quantize_stat] = flags.quantize_stat
if flags.model_data_format:
model_data_format = flags.model_data_format
else:
model_data_format = configs.get(YAMLKeyword.model_data_format, "file")
embed_model_data = model_data_format == ModelFormat.code
if flags.model_graph_format:
model_graph_format = flags.model_graph_format
else:
model_graph_format = configs.get(YAMLKeyword.model_graph_format,
"file")
embed_graph_def = model_graph_format == ModelFormat.code
if flags.enable_micro:
mace_check((not embed_model_data) and (not embed_graph_def),
ModuleName.YAML_CONFIG,
"You should specify file mode to convert micro model.")
if embed_graph_def:
os.makedirs(model_header_dir)
sh_commands.gen_mace_engine_factory_source(
configs[YAMLKeyword.models].keys(), embed_model_data)
sh.cp("-f", glob.glob("mace/codegen/engine/*.h"), model_header_dir)
convert.convert(configs, MODEL_CODEGEN_DIR, flags.enable_micro)
for model_name, model_config in configs[YAMLKeyword.models].items():
if flags.enable_micro:
data_type = model_config.get(YAMLKeyword.data_type, "")
mace_check(
data_type == FPDataType.fp32_fp32.value
or data_type == FPDataType.bf16_fp32.value,
ModuleName.YAML_CONFIG,
"You should specify fp32_fp32 or bf16_fp32 data type "
"to convert micro model.")
model_codegen_dir = "%s/%s" % (MODEL_CODEGEN_DIR, model_name)
encrypt.encrypt(
model_name, "%s/model/%s.pb" % (model_codegen_dir, model_name),
"%s/model/%s.data" % (model_codegen_dir, model_name),
config_parser.parse_device_type(model_config[YAMLKeyword.runtime]),
model_codegen_dir, bool(model_config.get(YAMLKeyword.obfuscate,
1)),
model_graph_format == "code", model_data_format == "code")
if model_graph_format == ModelFormat.file:
sh.mv("-f", '%s/model/%s.pb' % (model_codegen_dir, model_name),
model_output_dir)
sh.mv("-f", '%s/model/%s.data' % (model_codegen_dir, model_name),
model_output_dir)
if flags.enable_micro:
sh.mv(
"-f", '%s/model/%s_micro.tar.gz' %
(model_codegen_dir, model_name), model_output_dir)
else:
if not embed_model_data:
sh.mv("-f",
'%s/model/%s.data' % (model_codegen_dir, model_name),
model_output_dir)
sh.cp("-f", glob.glob("mace/codegen/models/*/code/*.h"),
model_header_dir)
MaceLogger.summary(
StringFormatter.block("Model %s converted" % model_name))
if model_graph_format == ModelFormat.code:
build_model_lib(configs, flags.address_sanitizer, flags.debug_mode)
print_library_summary(configs)
def convert_func(flags):
configs = config_parser.parse(flags.config)
print(configs)
library_name = configs[YAMLKeyword.library_name]
if not os.path.exists(BUILD_OUTPUT_DIR):
os.makedirs(BUILD_OUTPUT_DIR)
elif os.path.exists(os.path.join(BUILD_OUTPUT_DIR, library_name)):
sh.rm("-rf", os.path.join(BUILD_OUTPUT_DIR, library_name))
os.makedirs(os.path.join(BUILD_OUTPUT_DIR, library_name))
if not os.path.exists(BUILD_DOWNLOADS_DIR):
os.makedirs(BUILD_DOWNLOADS_DIR)
model_output_dir = \
'%s/%s/%s' % (BUILD_OUTPUT_DIR, library_name, MODEL_OUTPUT_DIR_NAME)
model_header_dir = \
'%s/%s/%s' % (BUILD_OUTPUT_DIR, library_name, MODEL_HEADER_DIR_PATH)
# clear output dir
if os.path.exists(model_output_dir):
sh.rm("-rf", model_output_dir)
os.makedirs(model_output_dir)
if os.path.exists(model_header_dir):
sh.rm("-rf", model_header_dir)
if os.path.exists(MODEL_CODEGEN_DIR):
sh.rm("-rf", MODEL_CODEGEN_DIR)
if os.path.exists(ENGINE_CODEGEN_DIR):
sh.rm("-rf", ENGINE_CODEGEN_DIR)
if flags.model_data_format:
model_data_format = flags.model_data_format
else:
model_data_format = configs.get(YAMLKeyword.model_data_format, "file")
embed_model_data = model_data_format == ModelFormat.code
if flags.model_graph_format:
model_graph_format = flags.model_graph_format
else:
model_graph_format = configs.get(YAMLKeyword.model_graph_format,
"file")
if model_graph_format == ModelFormat.code:
os.makedirs(model_header_dir)
sh_commands.gen_mace_engine_factory_source(
configs[YAMLKeyword.models].keys(), embed_model_data)
sh.cp("-f", glob.glob("mace/codegen/engine/*.h"), model_header_dir)
convert.convert(configs, MODEL_CODEGEN_DIR)
for model_name, model_config in configs[YAMLKeyword.models].items():
model_codegen_dir = "%s/%s" % (MODEL_CODEGEN_DIR, model_name)
encrypt.encrypt(
model_name, "%s/model/%s.pb" % (model_codegen_dir, model_name),
"%s/model/%s.data" % (model_codegen_dir, model_name),
config_parser.parse_device_type(model_config[YAMLKeyword.runtime]),
model_codegen_dir, bool(model_config.get(YAMLKeyword.obfuscate,
1)),
model_graph_format == "code", model_data_format == "code")
if model_graph_format == ModelFormat.file:
sh.mv("-f", '%s/model/%s.pb' % (model_codegen_dir, model_name),
model_output_dir)
sh.mv("-f", '%s/model/%s.data' % (model_codegen_dir, model_name),
model_output_dir)
else:
if not embed_model_data:
sh.mv("-f",
'%s/model/%s.data' % (model_codegen_dir, model_name),
model_output_dir)
sh.cp("-f", glob.glob("mace/codegen/models/*/code/*.h"),
model_header_dir)
MaceLogger.summary(
StringFormatter.block("Model %s converted" % model_name))
if model_graph_format == ModelFormat.code:
build_model_lib(configs, flags.address_sanitizer, flags.debug_mode)
print_library_summary(configs)
#
# graph = tf.Graph()
# with tf.compat.v1.Session(graph=graph) as sess:
# loader = tf.compat.v1.train.import_meta_graph(trained_checkpoint_prefix +
# '.meta')
# loader.restore(sess, trained_checkpoint_prefix)
#
# builder = tf.compat.v1.saved_model.builder.SavedModelBuilder(export_directory)
# builder.add_meta_graph_and_variables(sess,
# [tf.saved_model.TRAINING,
# tf.saved_model.SERVING],
# strip_default_attrs=True)
# builder.save()
model_config_file = "cfg/darknet.cfg"
parsed_config = config_parser.parse(model_config_file)
detection_model = Detector(parsed_config, 1)
detection_model = detection_model.build_model((416, 416, 3), 1)
detection_model.layers[0].trainable = False
checkpoint = tf.train.Checkpoint(model=detection_model)
latest_checkpoint_path = tf.train.latest_checkpoint("/home/karan/Checkpoint/")
checkpoint.restore(latest_checkpoint_path)
# shape = (1, 416, 416, 3)
# sample_image = np.random.randint(0, 255, size=(1, 416, 416, 3), dtype=np.uint8)
# input_image = np.array(sample_image / 255, dtype=np.float32)
# model = tf.keras.models.load_model("/home/karan/Checkpoint/")
# model.build_model(shape[1:], shape[0])
