# Set Training parameters
params = Tester.TestParams()
params.subnet_name = 'detection_subnet'
params.gpus = [0]
params.ckpt = './demo/models/ckpt_baseline_resnet101.h5'
params.batch_size = 25 * len(params.gpus)
params.print_freq = 100
# validation data
valid_data = get_loader(json_path,
data_dir,
mask_dir,
inp_size,
feat_stride,
preprocess='resnet',
batch_size=params.batch_size - 10 * len(params.gpus),
training=False,
shuffle=False,
num_workers=8,
subnet=params.subnet_name)
print('val dataset len: {}'.format(len(valid_data.dataset)))
# model
if backbone == 'resnet101':
model = poseNet(101)
elif backbone == 'resnet50':
model = poseNet(50)
for name, module in model.named_children():
for para in module.parameters():
if name in fpn_retinanet_para:
for para in module.parameters():
para.requires_grad = False
for name, module in model.named_children():
if name in retinanet_para:
for para in module.parameters():
para.requires_grad = False
for name, module in model.named_children():
if name in prn_para:
for para in module.parameters():
para.requires_grad = False
print("Loading dataset...")
# load training data
train_data = get_loader(json_path, data_dir,
mask_dir, inp_size, feat_stride,
'resnet', params.batch_size,
shuffle=True, training=True, num_workers=8)
print('train dataset len: {}'.format(len(train_data.dataset)))
# load validation data
valid_data = None
if params.val_nbatch > 0:
valid_data = get_loader(json_path, data_dir, mask_dir, inp_size,
feat_stride, preprocess='resnet', training=False,
batch_size=params.batch_size-2*len(params.gpus), shuffle=False, num_workers=8)
print('val dataset len: {}'.format(len(valid_data.dataset)))
trainable_vars = [param for param in model.parameters() if param.requires_grad]
if opt == 'adam':
print("Training with adam")
params.optimizer = torch.optim.Adam(