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,
checkpoint_name="checkpoints/pointnet2_cls",
best_name="checkpoints/pointnet2_cls_best",
lr_scheduler=lr_scheduler,
bnm_scheduler=bnm_scheduler,
viz=viz,
)
trainer.train(it,
start_epoch,
args.epochs,
train_loader,
test_loader,
best_loss=best_loss)
if start_epoch == args.epochs:
_ = trainer.eval_epoch(test_loader)
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,
checkpoint_name="checkpoints/"+os.path.basename(args.output),
best_name="checkpoints/"+os.path.basename(args.output)+"_best",
lr_scheduler=lr_scheduler,
bnm_scheduler=bnm_scheduler,
tensorboard_logger=log_name,
viz=viz,
)
trainer.train(
it, start_epoch, args.epochs, train_loader, val_loader, best_loss=best_loss
)
test_set = Indoor3DSemSeg(args.num_points,use_colour=args.colours, split="test")
test_loader = DataLoader(
test_set,
batch_size=1,
shuffle=True,
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,
checkpoint_name=os.path.join('checkpoints', args.logfile),
best_name=os.path.join('checkpoints', args.logfile) + '_best',
lr_scheduler=lr_scheduler,
bnm_scheduler=bnm_scheduler,
viz=viz,
)
trainer.train(it,
start_epoch,
args.epochs,
train_loader,
test_loader,
best_loss=best_loss)
if start_epoch == args.epochs:
_ = trainer.eval_epoch(test_loader)
model_fn = model_fn_decorator(nn.CrossEntropyLoss())
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,
checkpoint_name="checkpoints/pointnet2_smeseg",
best_name="checkpoints/poitnet2_semseg_best",
lr_scheduler=lr_scheduler,
bnm_scheduler=bnm_scheduler,
viz=viz)
trainer.train(it,
start_epoch,
args.epochs,
train_loader,
test_loader,
best_loss=best_loss)
if start_epoch == args.epochs:
_ = trainer.eval_epoch(test_loader)
pin_memory=True,
)
model = Pointnet(input_channels=0, num_classes=40, use_xyz=False, type_r=typer)
model.cuda()
optimizer = optim.Adam(
model.parameters()
)
# load status from checkpoint
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
model_fn = model_fn_decorator(nn.CrossEntropyLoss())
trainer = pt_utils.Trainer(
model,
model_fn,
optimizer
)
loss, stat = trainer.eval_epoch(test_loader)
acc_per_cls = stat['acc']
loss_per_cls = stat['loss']
print('loss:', loss)
print('Average acc:', torch.mean(torch.Tensor(acc_per_cls)).item())
