import sys
root_logger = logging.getLogger()
stdout_handler = logging.StreamHandler(sys.stdout)
root_logger.addHandler(stdout_handler)
root_logger.setLevel(logging.DEBUG)
if __name__ == "__main__":
# get sym
# Try different network 18, 50, 101 to find the best one
sym = get_resnet_model('pretrained_models/resnet-34', 0)
_, args_params, aux_params = mx.model.load_checkpoint('detect_full_scale', 574)
# get some input
# change it to the data rec you create, and modify the batch_size
train_data = get_iterator(path='train.rec', data_shape=(3, 224, 224), label_width=7*7*9, batch_size=32, shuffle=True)
val_data = get_iterator(path='val.rec', data_shape=(3, 224, 224), label_width=7*7*9, batch_size=32)
# allocate gpu/cpu mem to the sym
mod = mx.mod.Module(symbol=sym, context=mx.gpu(0))
print("LOOKS GOOD SO FAR")
# print metric design
def loss_metric(label, pred):
"""
label: np.array->(batch_size, 7,7,9)
predict: same as label
"""
label = label.reshape((-1, 7, 7, 9))
pred = pred.reshape((-1, 7, 7, 9))
pred_shift = (pred+1)/2
cl = label[:, :, :, 0]
if cpu == 1:
context = mx.cpu(0)
else:
context = mx.gpu(0)
# get sym
# Try different network 18, 50, 101 to find the best one
sym = get_resnet_model('pretrained_models/resnet-34', 0, lambda_noobj)
_, args_params, aux_params = mx.model.load_checkpoint(
'pretrained_models/resnet-34', 0)
# get some input
# change it to the data rec you create, and modify the batch_size
train_data = get_iterator(path=train_data_path,
data_shape=(3, 224, 224),
label_width=7 * 7 * 9,
batch_size=batch_size,
shuffle=True)
val_data = get_iterator(path=val_data_path,
data_shape=(3, 224, 224),
label_width=7 * 7 * 9,
batch_size=batch_size)
# allocate gpu/cpu mem to the sym
mod = mx.mod.Module(symbol=sym, context=context)
# setup metric
# metric = mx.metric.create(loss_metric, allow_extra_outputs=True)
tme = time.time()
# setup monitor for debugging
root_logger = logging.getLogger()
stdout_handler = logging.StreamHandler(sys.stdout)
root_logger.addHandler(stdout_handler)
root_logger.setLevel(logging.DEBUG)
if __name__ == "__main__":
tensorboard_use = True
# get sym
# Try different network 18, 50, 101 to find the best one
sym = get_resnet_model('pretrained_models/resnet-34', 0)
_, args_params, aux_params = mx.model.load_checkpoint('pretrained_models/resnet-34', 0)
# get some input
# change it to the data rec you create, and modify the batch_size
train_data = get_iterator(path='DATA_rec/cat.rec', data_shape=(3, 224, 224), label_width=7*7*5, batch_size=32, shuffle=True)
val_data = get_iterator(path='DATA_rec/cat.rec', data_shape=(3, 224, 224), label_width=7*7*5, batch_size=32)
# allocate gpu/cpu mem to the sym
mod = mx.mod.Module(symbol=sym, context=mx.gpu(0))
# # print metric design
def loss_metric(label, pred):
"""
label: np.array->(batch_size, 7,7,5)
predict: same as label
"""
label = label.reshape((-1, 7, 7, 5))
pred = pred.reshape((-1, 7, 7, 5))
pred_shift = (pred+1)/2
cl = label[:, :, :, 0]
stdout_handler = logging.StreamHandler(sys.stdout)
root_logger.addHandler(stdout_handler)
root_logger.setLevel(logging.DEBUG)
if __name__ == "__main__":
# get sym
# Try different network 18, 50, 101 to find the best one
sym = get_resnet_model_YoloV1('pretrained_models/resnet-34', 0)
#_, args_params, aux_params = mx.model.load_checkpoint('pretrained_models/resnet-34', 0)
_, args_params, aux_params = mx.model.load_checkpoint(
'models/drive_detect', 200)
# get some input
# change it to the data rec you create, and modify the batch_size
train_data = get_iterator(path='DATA_rec/drive_small.rec',
data_shape=(3, 224, 224),
label_width=7 * 7 * 9,
batch_size=2,
shuffle=True)
val_data = get_iterator(path='DATA_rec/drive_small.rec',
data_shape=(3, 224, 224),
label_width=7 * 7 * 9,
batch_size=2)
# allocate gpu/cpu mem to the sym
mod = mx.mod.Module(symbol=sym, context=mx.cpu(0))
# setup metric
# metric = mx.metric.create(loss_metric, allow_extra_outputs=True)
tme = time.time()
logtrain = LogMetricsCallback('logs/train_' + str(tme))
logtest = LogMetricsCallback('logs/val_' + str(tme))