def run_prediction_pickle():
from config_farm import configuration_10_160_17L_4scales_v1 as cfg
import mxnet
data_pickle_file_path = '../data_provider_farm/data_folder/data_list_brainwash_test.pkl'
from data_provider_farm.pickle_provider import PickleProvider
pickle_provider = PickleProvider(data_pickle_file_path)
positive_index = pickle_provider.positive_index
negative_index = pickle_provider.negative_index
all_index = positive_index
print("num of positive: %d\nnum of negative: %d" %
(len(positive_index), len(negative_index)))
# import random
# random.shuffle(all_index)
symbol_file_path = '../symbol_farm/symbol_10_160_17L_4scales_v1_deploy.json'
model_file_path = '../saved_model/configuration_10_160_17L_4scales_v1_2019-09-20-13-08-26/train_10_160_17L_4scales_v1_iter_800000.params'
my_predictor = Predict(
mxnet=mxnet,
symbol_file_path=symbol_file_path,
model_file_path=model_file_path,
ctx=mxnet.gpu(0),
receptive_field_list=cfg.param_receptive_field_list,
receptive_field_stride=cfg.param_receptive_field_stride,
bbox_small_list=cfg.param_bbox_small_list,
bbox_large_list=cfg.param_bbox_large_list,
receptive_field_center_start=cfg.param_receptive_field_center_start,
num_output_scales=cfg.param_num_output_scales)
for idx in all_index:
im, _, bboxes_gt = pickle_provider.read_by_index(idx)
bboxes = my_predictor.predict(im,
resize_scale=1,
score_threshold=0.5,
top_k=10000,
NMS_threshold=0.6)
for bbox in bboxes:
cv2.rectangle(im, (bbox[0], bbox[1]), (bbox[2], bbox[3]),
(0, 255, 0), 2)
cv2.imshow('im', im)
key = cv2.waitKey()
for n in range(self.num_output_scales):
data_batch.append_label(mask_batch_list[n])
data_batch.append_label(label_batch_list[n])
return data_batch
def get_batch_size(self):
return self.batch_size
if __name__ == '__main__':
from config_farm import configuration_30_320_20L_4scales_v1 as cfg
from data_provider_farm.pickle_provider import PickleProvider
import mxnet
train_data_provider = PickleProvider(cfg.param_trainset_pickle_file_path)
train_dataiter = Multithread_DataIter_for_CrossEntropy(
mxnet_module=mxnet,
num_threads=cfg.param_num_thread_train_dataiter,
data_provider=train_data_provider,
batch_size=cfg.param_train_batch_size,
enable_horizon_flip=cfg.param_enable_horizon_flip,
enable_vertical_flip=cfg.param_enable_vertical_flip,
enable_random_brightness=cfg.param_enable_random_brightness,
brightness_params=cfg.param_brightness_factors,
enable_random_saturation=cfg.param_enable_random_saturation,
saturation_params=cfg.param_saturation_factors,
enable_random_contrast=cfg.param_enable_random_contrast,
contrast_params=cfg.param_contrast_factors,
enable_blur=cfg.param_enable_blur,
blur_params=cfg.param_blur_factors,
def run():
logging_GOCD.init_logging(log_file_path=param_log_file_path,
log_file_mode=param_log_mode)
logging.info('Preparing before training.')
sys.path.append('..')
from symbol_farm import symbol_10_320_20L_5scales_v2 as net
net_symbol, data_names, label_names = net.get_net_symbol()
net_initializer = mxnet.initializer.Xavier()
logging.info('Get net symbol successfully.')
# -----------------------------------------------------------------------------------------------
# 构造dataiter
from data_provider_farm.pickle_provider import PickleProvider
from data_iterator_farm.multithread_dataiter_for_cross_entropy_v2 import Multithread_DataIter_for_CrossEntropy as DataIter
train_data_provider = PickleProvider(param_trainset_pickle_file_path)
train_dataiter = DataIter(
mxnet_module=mxnet,
num_threads=param_num_thread_train_dataiter,
data_provider=train_data_provider,
batch_size=param_train_batch_size,
enable_horizon_flip=param_enable_horizon_flip,
enable_vertical_flip=param_enable_vertical_flip,
enable_random_brightness=param_enable_random_brightness,
brightness_params=param_brightness_factors,
enable_random_saturation=param_enable_random_saturation,
saturation_params=param_saturation_factors,
enable_random_contrast=param_enable_random_contrast,
contrast_params=param_contrast_factors,
enable_blur=param_enable_blur,
blur_params=param_blur_factors,
blur_kernel_size_list=param_blur_kernel_size_list,
neg_image_ratio=param_neg_image_ratio,
num_image_channels=param_num_image_channel,
net_input_height=param_net_input_height,
net_input_width=param_net_input_width,
num_output_scales=param_num_output_scales,
receptive_field_list=param_receptive_field_list,
receptive_field_stride=param_receptive_field_stride,
feature_map_size_list=param_feature_map_size_list,
receptive_field_center_start=param_receptive_field_center_start,
bbox_small_list=param_bbox_small_list,
bbox_large_list=param_bbox_large_list,
bbox_small_gray_list=param_bbox_small_gray_list,
bbox_large_gray_list=param_bbox_large_gray_list,
num_output_channels=param_num_output_channels,
neg_image_resize_factor_interval=param_neg_image_resize_factor_interval
)
val_dataiter = None
if param_valset_pickle_file_path != '' and param_val_batch_size != 0 and param_num_val_loops != 0 and param_num_thread_val_dataiter != 0:
val_data_provider = PickleProvider(param_valset_pickle_file_path)
val_dataiter = DataIter(
mxnet_module=mxnet,
num_threads=param_num_thread_val_dataiter,
data_provider=val_data_provider,
batch_size=param_val_batch_size,
enable_horizon_flip=param_enable_horizon_flip,
enable_vertical_flip=param_enable_vertical_flip,
enable_random_brightness=param_enable_random_brightness,
brightness_params=param_brightness_factors,
enable_random_saturation=param_enable_random_saturation,
saturation_params=param_saturation_factors,
enable_random_contrast=param_enable_random_contrast,
contrast_params=param_contrast_factors,
enable_blur=param_enable_blur,
blur_params=param_blur_factors,
blur_kernel_size_list=param_blur_kernel_size_list,
neg_image_ratio=param_neg_image_ratio,
num_image_channels=param_num_image_channel,
net_input_height=param_net_input_height,
net_input_width=param_net_input_width,
num_output_scales=param_num_output_scales,
receptive_field_list=param_receptive_field_list,
receptive_field_stride=param_receptive_field_stride,
feature_map_size_list=param_feature_map_size_list,
receptive_field_center_start=param_receptive_field_center_start,
bbox_small_list=param_bbox_small_list,
bbox_large_list=param_bbox_large_list,
bbox_small_gray_list=param_bbox_small_gray_list,
bbox_large_gray_list=param_bbox_large_gray_list,
num_output_channels=param_num_output_channels,
neg_image_resize_factor_interval=
param_neg_image_resize_factor_interval)
# ---------------------------------------------------------------------------------------------
# 构造metric
from metric_farm.metric_default import Metric
train_metric = Metric(param_num_output_scales)
val_metric = None
if val_dataiter is not None:
val_metric = Metric(param_num_output_scales)
train_GOCD.start_train(
param_dict=param_dict,
mxnet_module=mxnet,
context=[mxnet.gpu(i) for i in param_GPU_idx_list],
train_dataiter=train_dataiter,
train_metric=train_metric,
train_metric_update_frequency=param_train_metric_update_frequency,
num_train_loops=param_num_train_loops,
val_dataiter=val_dataiter,
val_metric=val_metric,
num_val_loops=param_num_val_loops,
validation_interval=param_validation_interval,
optimizer_name=param_optimizer_name,
optimizer_params=param_optimizer_params,
net_symbol=net_symbol,
net_initializer=net_initializer,
net_data_names=data_names,
net_label_names=label_names,
pretrained_model_param_path=param_pretrained_model_param_path,
display_interval=param_display_interval,
save_prefix=param_save_prefix,
model_save_interval=param_model_save_interval,
start_index=param_start_index)
def run():
logging_GOCD.init_logging(log_file_path=param_log_file_path,
log_file_mode=param_log_mode)
logging.info('Preparing before training.')
sys.path.append('..')
from net_farm.naivenet import naivenet20
net = naivenet20()
net_initializer = 'default'
# if torch.cuda.is_available():
# net.cuda()
# cudnn.benchmark = True
torch.cuda.set_device(param_gpu_id_list[0])
net = net.cuda(param_gpu_id_list[0])
# construct the learning rate scheduler
param_optimizer = optim.SGD(net.parameters(),
lr=param_learning_rate,
momentum=param_momentum,
weight_decay=param_weight_decay)
# param_lr_scheduler = optim.lr_scheduler.StepLR(param_optimizer, 500000, 0.1)
param_lr_scheduler = optim.lr_scheduler.MultiStepLR(
param_optimizer,
milestones=param_scheduler_step_list,
gamma=param_scheduler_factor)
loss_criterion = cross_entropy_with_hnm_for_one_class_detection(
param_hnm_ratio, param_num_output_scales)
logging.info('Get net model successfully.')
# -------------------------------------------------------------------------
# init dataiter
from data_provider_farm.pickle_provider import PickleProvider
from data_iterator_farm.multithread_dataiter_for_cross_entropy_v2 import \
Multithread_DataIter_for_CrossEntropy as DataIter
train_data_provider = PickleProvider(param_trainset_pickle_file_path)
train_dataiter = DataIter(
torch_module=torch,
num_threads=param_num_thread_train_dataiter,
data_provider=train_data_provider,
batch_size=param_train_batch_size,
enable_horizon_flip=param_enable_horizon_flip,
enable_vertical_flip=param_enable_vertical_flip,
enable_random_brightness=param_enable_random_brightness,
brightness_params=param_brightness_factors,
enable_random_saturation=param_enable_random_saturation,
saturation_params=param_saturation_factors,
enable_random_contrast=param_enable_random_contrast,
contrast_params=param_contrast_factors,
enable_blur=param_enable_blur,
blur_params=param_blur_factors,
blur_kernel_size_list=param_blur_kernel_size_list,
neg_image_ratio=param_neg_image_ratio,
num_image_channels=param_num_image_channel,
net_input_height=param_net_input_height,
net_input_width=param_net_input_width,
num_output_scales=param_num_output_scales,
receptive_field_list=param_receptive_field_list,
receptive_field_stride=param_receptive_field_stride,
feature_map_size_list=param_feature_map_size_list,
receptive_field_center_start=param_receptive_field_center_start,
bbox_small_list=param_bbox_small_list,
bbox_large_list=param_bbox_large_list,
bbox_small_gray_list=param_bbox_small_gray_list,
bbox_large_gray_list=param_bbox_large_gray_list,
num_output_channels=param_num_output_channels,
neg_image_resize_factor_interval=param_neg_image_resize_factor_interval
)
val_dataiter = None
if param_valset_pickle_file_path != '' and param_val_batch_size != 0 and \
param_num_val_loops != 0 and param_num_thread_val_dataiter != 0:
val_data_provider = PickleProvider(param_valset_pickle_file_path)
val_dataiter = DataIter(
torch_module=torch,
num_threads=param_num_thread_val_dataiter,
data_provider=val_data_provider,
batch_size=param_val_batch_size,
enable_horizon_flip=param_enable_horizon_flip,
enable_vertical_flip=param_enable_vertical_flip,
enable_random_brightness=param_enable_random_brightness,
brightness_params=param_brightness_factors,
enable_random_saturation=param_enable_random_saturation,
saturation_params=param_saturation_factors,
enable_random_contrast=param_enable_random_contrast,
contrast_params=param_contrast_factors,
enable_blur=param_enable_blur,
blur_params=param_blur_factors,
blur_kernel_size_list=param_blur_kernel_size_list,
neg_image_ratio=param_neg_image_ratio,
num_image_channels=param_num_image_channel,
net_input_height=param_net_input_height,
net_input_width=param_net_input_width,
num_output_scales=param_num_output_scales,
receptive_field_list=param_receptive_field_list,
receptive_field_stride=param_receptive_field_stride,
feature_map_size_list=param_feature_map_size_list,
receptive_field_center_start=param_receptive_field_center_start,
bbox_small_list=param_bbox_small_list,
bbox_large_list=param_bbox_large_list,
bbox_small_gray_list=param_bbox_small_gray_list,
bbox_large_gray_list=param_bbox_large_gray_list,
num_output_channels=param_num_output_channels,
neg_image_resize_factor_interval=
param_neg_image_resize_factor_interval)
# -------------------------------------------------------------------------
# init metric
from metric_farm.metric_default import Metric
train_metric = Metric(param_num_output_scales)
val_metric = None
if val_dataiter is not None:
val_metric = Metric(param_num_output_scales)
train_GOCD.start_train(
param_dict=param_dict,
task_name=param_task_name,
torch_module=torch,
gpu_id_list=param_gpu_id_list,
train_dataiter=train_dataiter,
train_metric=train_metric,
train_metric_update_frequency=param_train_metric_update_frequency,
num_train_loops=param_num_train_loops,
val_dataiter=val_dataiter,
val_metric=val_metric,
num_val_loops=param_num_val_loops,
validation_interval=param_validation_interval,
optimizer=param_optimizer,
lr_scheduler=param_lr_scheduler,
net=net,
net_initializer=net_initializer,
loss_criterion=loss_criterion,
pretrained_model_param_path=param_pretrained_model_param_path,
display_interval=param_display_interval,
save_prefix=param_save_prefix,
model_save_interval=param_model_save_interval,
start_index=param_start_index)