bnm_scheduler = pt_utils.BNMomentumScheduler(model, bnm_lmbd)
start_epoch = 1
best_prec = 0
best_loss = 1e10
else:
start_epoch, best_loss = pt_utils.load_checkpoint(
model, optimizer, filename=args.checkpoint.split(".")[0])
lr_scheduler = lr_sched.LambdaLR(optimizer,
lr_lbmd,
last_epoch=start_epoch)
bnm_scheduler = pt_utils.BNMomentumScheduler(model,
bnm_lmbd,
last_epoch=start_epoch)
model_fn = model_fn_decorator(nn.CrossEntropyLoss())
# viz = pt_utils.VisdomViz(port=args.visdom_port)
# viz.text(str(vars(args)))
trainer = pt_utils.Trainer(
model,
model_fn,
optimizer,
checkpoint_name=
"/kitti_semantic/Pointnet2_PyTorch-master/pointnet2/train/checkpoints/pointnet2_smeseg",
best_name=
"/kitti_semantic/Pointnet2_PyTorch-master/pointnet2/train/checkpoints/poitnet2_semseg_best",
lr_scheduler=lr_scheduler,
bnm_scheduler=bnm_scheduler
# viz=viz
def test_xyz():
model = Pointnet2MSG(3, input_channels=3)
pytest.helpers.semseg_test_xyz(model,
model_fn_decorator(nn.CrossEntropyLoss()))
model, optimizer, filename=args.checkpoint.split(".")[0])
if checkpoint_status is not None:
it, start_epoch, best_loss = checkpoint_status
lr_scheduler = lr_sched.LambdaLR(optimizer,
lr_lambda=lr_lbmd,
last_epoch=it)
bnm_scheduler = pt_utils.BNMomentumScheduler(model,
bn_lambda=bnm_lmbd,
last_epoch=it)
print("Defined even more")
it = max(it, 0) # for the initialize value of `trainer.train`
weights = train_set.get_weights()
model_fn = model_fn_decorator(
nn.CrossEntropyLoss(weight=weights.float()).cuda())
args.visdom = False
if args.visdom:
viz = pt_utils.VisdomViz(port=args.visdom_port)
else:
viz = pt_utils.CmdLineViz()
print("initialized visdom (not)")
viz.text(pprint.pformat(vars(args)))
if not osp.isdir("checkpoints"):
os.makedirs("checkpoints")
trainer = pt_utils.Trainer(
model,
model_fn,
optimizer,
if args.checkpoint is not None:
checkpoint_status = pt_utils.load_checkpoint(
model, optimizer, filename=args.checkpoint.split(".")[0]
)
if checkpoint_status is not None:
it, start_epoch, best_loss = checkpoint_status
lr_scheduler = lr_sched.LambdaLR(optimizer, lr_lambda=lr_lbmd, last_epoch=it)
bnm_scheduler = pt_utils.BNMomentumScheduler(
model, bn_lambda=bnm_lmbd, last_epoch=it
)
it = max(it, 0) # for the initialize value of `trainer.train`
if args.weights!="":
weights=torch.from_numpy(np.loadtxt(args.weights)).float().cuda()
model_fn = model_fn_decorator(nn.CrossEntropyLoss(weight=weights))
else:
model_fn = model_fn_decorator(nn.CrossEntropyLoss(ignore_index=26))
if args.visdom:
viz = pt_utils.VisdomViz(port=args.visdom_port)
else:
viz = pt_utils.CmdLineViz()
viz.text(pprint.pformat(vars(args)))
if not osp.isdir("checkpoints"):
os.makedirs("checkpoints")
trainer = pt_utils.Trainer(
model,
checkpoint_status = pt_utils.load_checkpoint(
model, optimizer, filename=args.checkpoint.split(".")[0])
if checkpoint_status is not None:
it, start_epoch, best_loss = checkpoint_status
lr_scheduler = lr_sched.LambdaLR(optimizer,
lr_lambda=lr_lbmd,
last_epoch=it)
bnm_scheduler = pt_utils.BNMomentumScheduler(model,
bn_lambda=bnm_lmbd,
last_epoch=it)
it = max(it, 0) # for the initialize value of `trainer.train`
# model_fn = model_fn_decorator(nn.CrossEntropyLoss())
model_fn = model_fn_decorator(focal_loss.FocalLoss(class_num=2, alpha=torch.cuda.FloatTensor([1.0, 3.0]), \
gamma=2))
if args.visdom:
viz = pt_utils.VisdomViz(port=args.visdom_port)
else:
viz = pt_utils.CmdLineViz()
viz.text(pprint.pformat(vars(args)))
if not osp.isdir("checkpoints"):
os.makedirs("checkpoints")
trainer = pt_utils.Trainer(
model,
model_fn,
optimizer,
