def main(hparams):
"""
Main training routine specific for this project
:param hparams:
"""
cfg = get_cfg()
cfg = add_config(cfg)
cfg.setup_cfg_with_hparams(hparams)
if hasattr(hparams, "test_only") and hparams.test_only:
model = build_module(cfg)
trainer = build_trainer(cfg, hparams)
trainer.test(model)
else:
model = build_module(cfg)
trainer = build_trainer(cfg, hparams)
trainer.fit(model)
import sys
sys.path.append('..')
import torchline as tl
from config.config import add_config
import models
import torch
cfg = tl.config.get_cfg()
cfg = add_config(cfg)
cfg.merge_from_file('../config/config.yml')
cfg.model.n_input_channels = 1
x = torch.rand(1, 1, 32, 64, 64)
model_names = {
'mc3_18': [],
'r3d_18': [],
'r2plus1d_18': [],
'densenet3d': [121, 169, 201, 264],
'resnet3d': [10, 18, 34, 50, 101, 152, 200],
# 'wide_resnet3d': [50,101,152,200],
# 'resnext3d': [50,101,152,200],
# 'preact_resnet3d': [10,18,34,50,101,152,200],
}
for name in model_names:
depths = model_names[name]
if name == 'densenet3d':
cfg.model.model_depth = 121
if name in ['resnet3d', 'wide_resnet3d', 'resnext3d', 'preact_resnet3d']:
cfg.model.model_depth = 50