def train_sequence_si():
ks = [10, 20, 30]
fecs = [[0, 1, 2, 3, 4, 5], [0, 1, 2, 3, 4, 5], [0, 1, 2, 3, 4, 5]]
logdir = 'modeldir/stage_all/seq_si/'
from loss import FECLoss
for i, k in enumerate(ks):
for fec in fecs[i]:
cfg = _allrun_config_si(k)
cfg['lr'] = 0.00001
cfg['batch'] = 64
cfg['epochs'] = 200
cfg['scheduler'] = True
cfg['patience'] = 30
cfg['step'] = 0
if fec % 2 == 1:
continue
if fec == 0:
cfg['criterion'] = torch.nn.BCELoss()
cfg['model'] = 'resnet18b4_si_k%d' % (k)
else:
cfg['criterion'] = FECLoss(alpha=cfg['batch'] * fec)
cfg['model'] = 'resnet18b4_si_k%d_fec%d' % (k, fec)
cfg['model_dir'] = '%s/%s' % (logdir, cfg['model'])
model = nn.DataParallel(sinet.SiNet(nblock=4, k=k).cuda())
train.run_train(model, cfg)
def train_resnet_pj(s=2, k=10):
cfg = util.default_cfg()
cfg = train._config_pj_dataset(cfg, s, k)
cfg['model'] = 'resnet_pj_k%d' % (k)
cfg['model_dir'] = 'modeldir/stage%d/resnet_pj_k%d' % (s, k)
model = nn.DataParallel(sinet.SiNet(nblock=4, k=k).cuda())
cfg = train._train_config_pj(model, cfg)
train.run_train(model, cfg)
def train_resnet_si(s=2, k=10, val_index=4):
cfg = util.default_cfg()
cfg = train._config_si_dataset(cfg, s, k)
cfg['model'] = 'resnet_si_k%d_val%d' % (k, val_index)
cfg['model_dir'] = 'modeldir/stage%d/resnet_si_k%d_val%d' % (s, k,
val_index)
model = nn.DataParallel(sinet.SiNet(nblock=4, k=k).cuda())
cfg = train._train_config_si(model, cfg)
train.run_train(model, cfg)
def train_resnet_stratify_si(s=2, k=10):
cfg = util.default_cfg()
cfg = train._config_stratify_si_dataset(cfg, s, k)
# from loss import FECLoss
# cfg['criterion'] = FECLoss(alpha=48)
# from loss import SFocalLoss
# cfg['criterion'] = SFocalLoss(gamma=1)
model = nn.DataParallel(sinet.SiNet(nblock=2, k=k).cuda())
cfg['model'] = 'resnet18b2_si_k%d' % (k)
cfg['model_dir'] = 'modeldir/stage%d/resnet18b2_si_k%d' % (s, k)
cfg = train._train_config_si(model, cfg)
cfg['scheduler'] = False
train.run_train(model, cfg)
def train_resnet_si(k=10):
cfg = _allrun_config_si(k)
from loss import FECLoss
cfg['criterion'] = FECLoss(alpha=64)
model = nn.DataParallel(sinet.SiNet(nblock=4, k=k).cuda())
cfg['model'] = 'resnet18b4_si_k%d_fec1' % (k)
cfg['model_dir'] = 'modeldir/stage_all/resnet18b4_si_k%d_fec1' % (k)
cfg['lr'] = 0.0001
model_pth = os.path.join(cfg['model_dir'], 'model.pth')
if os.path.exists(model_pth):
ckp = torch.load(model_pth)
model.load_state_dict(ckp['model'])
cfg['step'] = ckp['epoch'] + 1
print("load pretrained model", model_pth, "start epoch:", cfg['step'])
train.run_train(model, cfg)
def run_kfold_test(k=10):
for s in [6, 5, 4, 3, 2]:
test_dataset = dataset.SIDataset(mode='test', stage=s, k=k)
s_dir = 'modeldir/stage%d' % s
s_score = []
s_label = []
for val_index in [4, 3, 2, 1, 0]:
m_dir = '%s/resnet_si_k%d_val%d' % (s_dir, k, val_index)
model_pth = os.path.join(m_dir, 'model.pth')
model = nn.DataParallel(sinet.SiNet(nblock=4, k=k).cuda())
ckp = torch.load(model_pth)
model.load_state_dict(ckp['model'])
cfg = util.default_cfg()
cfg['test'] = test_dataset
cfg['batch'] = 128
cfg['collate'] = default_collate
cfg['instance'] = train._train_si
cfg['model'] = m_dir
np_score, np_label = train.run_test_score(model, cfg)
s_score.append(np_score)
s_label.append(np_label)
m_score = np.mean(np.stack(s_score, axis=0), axis=0)
print("m_score", m_score.shape, 'np_label', np_label.shape)
np_pd = (m_score > 0.5).astype(np.int)
mean_dir = '%s/resnet_si_k%d_mean' % (s_dir, k)
if not os.path.exists(mean_dir):
os.mkdir(mean_dir)
pth = os.path.join(mean_dir, 'metrics.csv')
util.write_metrics(pth, np_label, np_pd, np_score)