2018-09-18 15:47:38 +08:00
|
|
|
# encoding: utf-8
|
|
|
|
"""
|
|
|
|
@author: sherlock
|
|
|
|
@contact: sherlockliao01@gmail.com
|
|
|
|
"""
|
|
|
|
|
|
|
|
import argparse
|
|
|
|
import sys
|
2019-04-21 13:38:55 +08:00
|
|
|
from bisect import bisect_right
|
2018-09-18 15:47:38 +08:00
|
|
|
|
2018-10-18 19:04:28 +08:00
|
|
|
from torch.backends import cudnn
|
2018-09-18 15:47:38 +08:00
|
|
|
|
2019-01-10 18:39:31 +08:00
|
|
|
sys.path.append('.')
|
|
|
|
from config import cfg
|
2019-04-21 13:38:55 +08:00
|
|
|
from data import get_data_bunch
|
2019-01-10 18:39:31 +08:00
|
|
|
from engine.trainer import do_train
|
|
|
|
from layers import make_loss
|
2019-04-21 13:38:55 +08:00
|
|
|
from modeling import build_model
|
|
|
|
from utils.logger import Logger
|
|
|
|
from fastai.vision import *
|
2019-01-10 18:39:31 +08:00
|
|
|
|
|
|
|
|
2019-04-23 12:56:38 +08:00
|
|
|
def train(cfg, log_path):
|
2019-01-10 18:39:31 +08:00
|
|
|
# prepare dataset
|
2019-04-21 13:38:55 +08:00
|
|
|
data_bunch, test_labels, num_query = get_data_bunch(cfg)
|
|
|
|
|
2019-01-10 18:39:31 +08:00
|
|
|
# prepare model
|
2019-04-21 13:38:55 +08:00
|
|
|
model = build_model(cfg, data_bunch.c)
|
2019-01-10 18:39:31 +08:00
|
|
|
|
2019-04-21 13:38:55 +08:00
|
|
|
opt_func = partial(torch.optim.Adam)
|
2019-01-10 18:39:31 +08:00
|
|
|
|
2019-04-21 13:38:55 +08:00
|
|
|
def warmup_multistep(start: float, end: float, pct: float) -> float:
|
|
|
|
warmup_factor = 1
|
|
|
|
gamma = cfg.SOLVER.GAMMA
|
|
|
|
milestones = [1.0 * s / cfg.SOLVER.MAX_EPOCHS for s in cfg.SOLVER.STEPS]
|
|
|
|
warmup_iter = 1.0 * cfg.SOLVER.WARMUP_ITERS / cfg.SOLVER.MAX_EPOCHS
|
|
|
|
if pct < warmup_iter:
|
|
|
|
alpha = pct / warmup_iter
|
|
|
|
warmup_factor = cfg.SOLVER.WARMUP_FACTOR * (1 - alpha) + alpha
|
|
|
|
return start * warmup_factor * gamma ** bisect_right(milestones, pct)
|
|
|
|
|
|
|
|
lr_sched = Scheduler((cfg.SOLVER.BASE_LR, 0), cfg.SOLVER.MAX_EPOCHS, warmup_multistep)
|
2019-01-10 18:39:31 +08:00
|
|
|
|
2019-04-21 13:38:55 +08:00
|
|
|
loss_func = make_loss(cfg)
|
2019-01-10 18:39:31 +08:00
|
|
|
|
|
|
|
do_train(
|
|
|
|
cfg,
|
2019-04-23 12:56:38 +08:00
|
|
|
log_path,
|
2019-01-10 18:39:31 +08:00
|
|
|
model,
|
2019-04-21 13:38:55 +08:00
|
|
|
data_bunch,
|
|
|
|
test_labels,
|
|
|
|
opt_func,
|
|
|
|
lr_sched,
|
2019-01-10 18:39:31 +08:00
|
|
|
loss_func,
|
|
|
|
num_query
|
|
|
|
)
|
2018-09-18 15:47:38 +08:00
|
|
|
|
|
|
|
|
|
|
|
def main():
|
2019-01-10 18:39:31 +08:00
|
|
|
parser = argparse.ArgumentParser(description="ReID Baseline Training")
|
|
|
|
parser.add_argument(
|
|
|
|
"--config_file", default="", help="path to config file", type=str
|
|
|
|
)
|
|
|
|
parser.add_argument("opts", help="Modify config options using the command-line", default=None,
|
|
|
|
nargs=argparse.REMAINDER)
|
2018-09-18 15:47:38 +08:00
|
|
|
|
|
|
|
args = parser.parse_args()
|
2019-01-10 18:39:31 +08:00
|
|
|
|
|
|
|
num_gpus = int(os.environ["WORLD_SIZE"]) if "WORLD_SIZE" in os.environ else 1
|
|
|
|
|
|
|
|
if args.config_file != "":
|
|
|
|
cfg.merge_from_file(args.config_file)
|
|
|
|
cfg.merge_from_list(args.opts)
|
|
|
|
cfg.freeze()
|
|
|
|
|
2019-04-23 12:56:38 +08:00
|
|
|
log_path = Path(os.path.join(cfg.OUTPUT_DIR, 'log.txt'))
|
|
|
|
with log_path.open('w') as f: f.write('{}\n'.format(args))
|
2019-04-21 13:38:55 +08:00
|
|
|
print(args)
|
2019-01-10 18:39:31 +08:00
|
|
|
|
|
|
|
if args.config_file != "":
|
2019-04-21 13:38:55 +08:00
|
|
|
print("Loaded configuration file {}".format(args.config_file))
|
2019-01-10 18:39:31 +08:00
|
|
|
with open(args.config_file, 'r') as cf:
|
|
|
|
config_str = "\n" + cf.read()
|
2019-04-21 13:38:55 +08:00
|
|
|
print(config_str)
|
|
|
|
print("Running with config:\n{}".format(cfg))
|
2019-04-23 12:56:38 +08:00
|
|
|
with log_path.open('a') as f: f.write('{}\n'.format(cfg))
|
2018-10-18 19:04:28 +08:00
|
|
|
cudnn.benchmark = True
|
2019-04-23 12:56:38 +08:00
|
|
|
train(cfg, log_path)
|
|
|
|
|
2018-09-18 15:47:38 +08:00
|
|
|
|
|
|
|
|
|
|
|
if __name__ == '__main__':
|
|
|
|
main()
|