def main():
# argument parse and create log
args = parse_args()
cfg.set_args(args.gpu_ids, args.mode, args.continue_train)
cudnn.benchmark = True
trainer = Trainer()
trainer._make_batch_generator()
trainer._make_model()
# train
for epoch in range(trainer.start_epoch, cfg.end_epoch):
trainer.set_lr(epoch)
trainer.tot_timer.tic()
trainer.read_timer.tic()
for itr, (inputs, targets,
meta_info) in enumerate(trainer.batch_generator):
trainer.read_timer.toc()
trainer.gpu_timer.tic()
# forward
trainer.optimizer.zero_grad()
loss = trainer.model(inputs, targets, meta_info, 'train')
loss = {k: loss[k].mean() for k in loss}
# backward
sum(loss[k] for k in loss).backward()
trainer.optimizer.step()
trainer.gpu_timer.toc()
screen = [
'Epoch %d/%d itr %d/%d:' %
(epoch, cfg.end_epoch, itr, trainer.itr_per_epoch),
'lr: %g' % (trainer.get_lr()),
'speed: %.2f(%.2fs r%.2f)s/itr' %
(trainer.tot_timer.average_time,
trainer.gpu_timer.average_time,
trainer.read_timer.average_time),
'%.2fh/epoch' % (trainer.tot_timer.average_time / 3600. *
trainer.itr_per_epoch),
]
screen += [
'%s: %.4f' % ('loss_' + k, v.detach())
for k, v in loss.items()
]
trainer.logger.info(' '.join(screen))
trainer.tot_timer.toc()
trainer.tot_timer.tic()
trainer.read_timer.tic()
trainer.save_model(
{
'epoch': epoch,
'network': trainer.model.state_dict(),
'optimizer': trainer.optimizer.state_dict(),
}, epoch)
def main():
# argument parse and create log
args = parse_args()
cfg.set_args(args.gpu_ids, args.continue_train)
cudnn.fastest = True
cudnn.benchmark = True
trainer = Trainer()
trainer._make_batch_generator()
trainer._make_model()
# train
for epoch in range(trainer.start_epoch, cfg.end_epoch):
trainer.set_lr(epoch)
trainer.tot_timer.tic()
trainer.read_timer.tic()
for itr, (input_img, k_value, root_img, root_vis, joints_have_depth) in enumerate(trainer.batch_generator):
trainer.read_timer.toc()
trainer.gpu_timer.tic()
# forward
trainer.optimizer.zero_grad()
target = {'coord': root_img, 'vis': root_vis, 'have_depth': joints_have_depth}
loss_coord = trainer.model(input_img, k_value, target)
loss_coord = loss_coord.mean();
# backward
loss = loss_coord
loss.backward()
trainer.optimizer.step()
trainer.gpu_timer.toc()
screen = [
'Epoch %d/%d itr %d/%d:' % (epoch, cfg.end_epoch, itr, trainer.itr_per_epoch),
'lr: %g' % (trainer.get_lr()),
'speed: %.2f(%.2fs r%.2f)s/itr' % (
trainer.tot_timer.average_time, trainer.gpu_timer.average_time, trainer.read_timer.average_time),
'%.2fh/epoch' % (trainer.tot_timer.average_time / 3600. * trainer.itr_per_epoch),
'%s: %.4f' % ('loss_coord', loss_coord.detach()),
]
trainer.logger.info(' '.join(screen))
trainer.tot_timer.toc()
trainer.tot_timer.tic()
trainer.read_timer.tic()
trainer.save_model({
'epoch': epoch,
'network': trainer.model.state_dict(),
'optimizer': trainer.optimizer.state_dict(),
}, epoch)
def main():
# argument parse and create log
args = parse_args()
# restrict one gpu : not support distributed learning
cfg.set_args(args.gpu)
cudnn.benchmark = True
# set trainer
trainer = Trainer()
trainer.build_dataloader()
trainer.build_model()
trainer.set_optimizer()
trainer.set_scheduler()
start_epoch = 0
# load model
if cfg.load_checkpoint:
start_epoch = trainer.load_model()
# logger
logger = Logger()
# train model
for epoch in range(start_epoch, cfg.epoch):
for i, data in enumerate(trainer.dataloader):
trainer.optimizer.zero_grad()
proposal_loss, detection_loss = trainer.model(data)
loss = proposal_loss[
0] + proposal_loss[1] * cfg.pro_loc_lambda + detection_loss[
0] + detection_loss[1] * cfg.det_loc_lambda
loss.backward()
trainer.optimizer.step()
logger.log(proposal_loss, detection_loss, epoch, i,
epoch * len(trainer.dataloader) + i)
if cfg.visualize_switch:
if i % 500 == 0:
cfg.visualize = True
else:
cfg.visualize = False
trainer.scheduler.step()
if cfg.save_checkpoint:
trainer.save_model(epoch)
def main():
# argument parse and create log
args = parse_args()
cfg.set_args(args.gpu_ids, args.continue_train)
cudnn.fastest = True
cudnn.benchmark = True
trainer = Trainer()
trainer._make_batch_generator()
trainer._make_model()
# train
for epoch in range(trainer.start_epoch, cfg.end_epoch):
trainer.set_lr(epoch)
trainer.tot_timer.tic()
trainer.read_timer.tic()
for itr in range(trainer.itr_per_epoch):
input_img_list, joint_img_list, joint_vis_list, joints_have_depth_list = [], [], [], []
for i in range(len(cfg.trainset)):
try:
input_img, joint_img, joint_vis, joints_have_depth = next(trainer.iterator[i])
except StopIteration:
trainer.iterator[i] = iter(trainer.batch_generator[i])
input_img, joint_img, joint_vis, joints_have_depth = next(trainer.iterator[i])
input_img_list.append(input_img)
joint_img_list.append(joint_img)
joint_vis_list.append(joint_vis)
joints_have_depth_list.append(joints_have_depth)
# aggregate items from different datasets into one single batch
input_img = torch.cat(input_img_list,dim=0)
joint_img = torch.cat(joint_img_list,dim=0)
joint_vis = torch.cat(joint_vis_list,dim=0)
joints_have_depth = torch.cat(joints_have_depth_list,dim=0)
# shuffle items from different datasets
rand_idx = []
for i in range(len(cfg.trainset)):
rand_idx.append(torch.arange(i,input_img.shape[0],len(cfg.trainset)))
rand_idx = torch.cat(rand_idx,dim=0)
rand_idx = rand_idx[torch.randperm(input_img.shape[0])]
input_img = input_img[rand_idx]; joint_img = joint_img[rand_idx]; joint_vis = joint_vis[rand_idx]; joints_have_depth = joints_have_depth[rand_idx];
target = {'coord': joint_img, 'vis': joint_vis, 'have_depth': joints_have_depth}
trainer.read_timer.toc()
trainer.gpu_timer.tic()
trainer.optimizer.zero_grad()
# forward
loss_coord = trainer.model(input_img, target)
loss_coord = loss_coord.mean()
# backward
loss = loss_coord
loss.backward()
trainer.optimizer.step()
trainer.gpu_timer.toc()
screen = [
'Epoch %d/%d itr %d/%d:' % (epoch, cfg.end_epoch, itr, trainer.itr_per_epoch),
'lr: %g' % (trainer.get_lr()),
'speed: %.2f(%.2fs r%.2f)s/itr' % (
trainer.tot_timer.average_time, trainer.gpu_timer.average_time, trainer.read_timer.average_time),
'%.2fh/epoch' % (trainer.tot_timer.average_time / 3600. * trainer.itr_per_epoch),
'%s: %.4f' % ('loss_coord', loss_coord.detach()),
]
trainer.logger.info(' '.join(screen))
trainer.tot_timer.toc()
trainer.tot_timer.tic()
trainer.read_timer.tic()
trainer.save_model({
'epoch': epoch,
'network': trainer.model.state_dict(),
'optimizer': trainer.optimizer.state_dict(),
}, epoch)
def main():
# argument parse and create log
args = parse_args()
cfg.set_args(args.gpu_ids, args.continue_train)
cudnn.fastest = True
cudnn.benchmark = True
cudnn.deterministic = False
cudnn.enabled = True
trainer = Trainer(cfg)
trainer._make_batch_generator()
trainer._make_model()
# train
for epoch in range(trainer.start_epoch, cfg.end_epoch):
trainer.scheduler.step()
trainer.tot_timer.tic()
trainer.read_timer.tic()
for itr, (input_img, joint_img, joint_vis,
joints_have_depth) in enumerate(trainer.batch_generator):
trainer.read_timer.toc()
trainer.gpu_timer.tic()
trainer.optimizer.zero_grad()
input_img = input_img.cuda()
joint_img = joint_img.cuda()
joint_vis = joint_vis.cuda()
joints_have_depth = joints_have_depth.cuda()
# forward
heatmap_out = trainer.model(input_img)
# backward
JointLocationLoss = trainer.JointLocationLoss(
heatmap_out, joint_img, joint_vis, joints_have_depth)
loss = JointLocationLoss
loss.backward()
trainer.optimizer.step()
trainer.gpu_timer.toc()
screen = [
'Epoch %d/%d itr %d/%d:' %
(epoch, cfg.end_epoch, itr, trainer.itr_per_epoch),
'lr: %g' % (trainer.scheduler.get_lr()[0]),
'speed: %.2f(%.2fs r%.2f)s/itr' %
(trainer.tot_timer.average_time,
trainer.gpu_timer.average_time,
trainer.read_timer.average_time),
'%.2fh/epoch' % (trainer.tot_timer.average_time / 3600. *
trainer.itr_per_epoch),
'%s: %.4f' % ('loss_loc', JointLocationLoss.detach()),
]
trainer.logger.info(' '.join(screen))
trainer.tot_timer.toc()
trainer.tot_timer.tic()
trainer.read_timer.tic()
trainer.save_model(
{
'epoch': epoch,
'network': trainer.model.state_dict(),
'optimizer': trainer.optimizer.state_dict(),
'scheduler': trainer.scheduler.state_dict(),
}, epoch)