def main():
train = Server(GetConfig("Canonical"), COCOSourceConfig("../dataset/coco_train_dataset.h5"), 5555, "Train", shuffle=True, augment=True)
val = Server(GetConfig("Canonical"), COCOSourceConfig("../dataset/coco_val_dataset.h5"), 5556, "Val", shuffle=False, augment=False)
processes = [train, val]
while None in [p.process.exitcode for p in processes]:
print("exitcodes", [p.process.exitcode for p in processes])
for p in processes:
if p.process.exitcode is None:
p.join()
from addins.head_counter_config import HeadCounterConfig
from model import get_testing_model
from glob import glob
toloka_dir = os.path.abspath(
'/opt/home/anatolix/YaDisk/Приложения/Яндекс.Толока')
results_dir_mask = os.path.join(toloka_dir,
"Results/Heads/assignments_*.tsv.csv")
img_dir = os.path.join(toloka_dir, "Yandex.Toloka/pochta")
task = sys.argv[1]
assert task == "predict" or task == "render"
config_name = sys.argv[2]
model = sys.argv[3]
config = GetConfig(config_name)
val_size = 19
val_ids = None
train_ids = None
def prepare(config, model_file):
model = get_testing_model(np_branch1=config.paf_layers,
np_branch2=config.heat_layers + 1)
print("using model:", model_file)
model.load_weights(model_file)
def db_release():
requests.request('GET', 'http://127.0.0.1:5000/shutdown')
conf = GetConfig()
conf.release_db_disk(logger)
start_server()
def clear_dir():
conf = GetConfig()
conf.clean_tmp_file()
for i in range(matches.shape[0]):
okp_threshold = eval_result.params.iouThrs[i]
scores['matched_%.2f' % okp_threshold] = sum(matches[i, :] != 0)
scores['average'] = np.mean(np.sum(matches != 0, axis=1)) / scores['gt_person_count']
return scores
evalImgs = eval_result.evalImgs
scores = [convert_match_to_score(image_match) for image_match in evalImgs if image_match is not None]
return pd.DataFrame(scores)
if __name__ == "__main__":
config = GetConfig('Canonical')
params, model_params = config_reader(figstr='testing/config')
from posenet.mymodel3 import get_testing_model
# with tf.device("/cpu:0"):
# model_single = get_testing_model(np_branch1=config.paf_layers, np_branch2=config.heat_layers, stages=1)
# model = multi_gpu_model(model_single, gpus=2)
model = get_testing_model(np_branch1=config.paf_layers, np_branch2=config.heat_layers+1, stages=3) # fixme: background + 1
training_dir = './training/'
trained_models = [
'weights'
# 'weights-cpp-lr'
# 'weights-python-last',
]
optimal_epoch_loss = 'val_weight_stage6_L1_loss'
trans_status = True
os.remove(tp)
if f_type == 'video':
obj_id = self.media_shortcut(tmp)
info = {'shortcut': obj_id if obj_id else ''}
if f_type == 'audio':
info = self.media_info_extract(tmp)
res.append({
'file_path': tmp,
'des': info,
"trans_status": trans_status
})
return config.build_response('after_upload', res, 200)
config = GetConfig()
fo = FileOperation()
file = Blueprint('file', __name__)
globalLock = threading.Lock()
@file.route('/upload', methods=['POST'])
def file_upload():
if request.method == 'POST':
if request.form["name"] and request.form['f_type'] and request.form[
'unique_id']:
tmp_dir = config.get_tmp_by_type(
request.form['f_type']) + '\\' + request.form['unique_id']
globalLock.acquire()
if not os.path.exists(tmp_dir):
os.mkdir(tmp_dir)
def _get_files(self):
return GetConfig().get_log()
from keras.utils import plot_model
from keras import Model, Sequential
from keras.layers import Input, ZeroPadding2D, BatchNormalization, Activation, Conv2D, MaxPooling2D, \
AveragePooling2D, UpSampling2D, Lambda, Dropout
from keras.layers.merge import Concatenate, Multiply, Add
from keras.regularizers import l2
from keras.initializers import random_normal, constant
from keras.applications.vgg19 import preprocess_input
import keras.backend as K
import tensorflow as tf
import re
import numpy as np
from config import COCOSourceConfig, GetConfig
config = GetConfig("Canonical")
def relu(x):
return Activation('relu')(x)
def conv(x,
nf,
ks,
name,
weight_decay,
stride=1,
use_bias=False,
use_bn=False,
use_relu=True):
kernel_reg = l2(weight_decay[0]) if weight_decay else None
if self.mask_cache is not None:
return self.mask_cache
head_layer = global_config.find_heat_layer('HeadCenter')
if global_config.num_parts == 1:
#background is ok
print("Layers will be kept: ", head_layer, "background")
self.mask_cache = np.ones(global_config.parts_shape,
dtype=np.float)
else:
print("Layers will be kept: ", head_layer)
self.mask_cache = np.zeros(global_config.parts_shape,
dtype=np.float)
self.mask_cache[:, :, head_layer] = 1.
return self.mask_cache
def source(self):
return self.hdf5_source
Configs["HeadCount"] = HeadCounterConfig
Configs["HeadTrim"] = HeadCounterTrimmedConfig
if __name__ == "__main__":
# test it
foo = GetConfig("HeadCount")