def eval(cfg, model, loader, criterion, publisher="test"):
model.eval()
# metrics
acc_meter = MultiAssessmentMeter(num_classes=cfg.dataset.num_classes,
metrics=["class", "overall", "iou"])
batch_loss = LossMeter()
meters = (acc_meter, batch_loss)
with torch.no_grad():
for _, data in enumerate(loader):
_ = processing(model, criterion, data, meters, cfg.general.device)
# get epoch loss and accuracy
epoch_loss = batch_loss.compute()
epoch_acc = acc_meter.compute()
# save loss and acc to tensorboard
log_dict = {
"{}/loss".format(publisher): epoch_loss,
"{}/mAcc".format(publisher): epoch_acc["class"],
"{}/oAcc".format(publisher): epoch_acc["overall"],
"{}/IoU".format(publisher): epoch_acc["iou"]
}
return log_dict
def train(cfg,
model,
dataset,
optimizer,
criterion,
scheduler,
publisher="train"):
model.train()
loader = DataLoader(
#Subset(dataset["train"],range(320)),
dataset,
batch_size=cfg.general.batch_size,
num_workers=cfg.loader.nworkers,
pin_memory=True,
shuffle=True,
)
# loader = tqdm(loader, ncols=100, desc=publisher)
acc_meter = MultiAssessmentMeter(num_classes=dataset.num_classes,
metrics=["class", "overall", "iou"])
batch_loss = LossMeter()
meters = (acc_meter, batch_loss)
for data in loader:
optimizer.zero_grad()
loss = processing(model, criterion, data, meters, cfg.general.device)
loss.backward()
optimizer.step()
# scheduler.step(data[0].shape[0]) # for TF ExponentialDecay
# nan
if torch.isnan(loss):
print("Train loss is nan.")
exit()
# update lr step
# scheduler.step()
# get epoch loss and acc
train_loss = batch_loss.compute()
train_acc = acc_meter.compute()
# save loss and acc to tensorboard
lr = scheduler.get_last_lr()[0]
log_dict = {
"lr": lr,
"train/loss": train_loss,
"train/mAcc": train_acc["class"],
"train/oAcc": train_acc["overall"],
"train/IoU": train_acc["iou"]
}
return log_dict
def train(cfg, model, dataset, optimizer, criterion, scheduler, publisher="train"):
model.train()
loader = DataLoader(
#Subset(dataset["train"],range(320)),
dataset,
batch_size=cfg.batch_size,
num_workers=cfg.nworkers,
pin_memory=True,
shuffle=True,
collate_fn=rotation_and_jitter
)
loader = tqdm(loader, ncols=100, desc=publisher)
acc_meter = MultiAssessmentMeter(num_classes=cfg.num_classes,
metrics=["class","overall"])
batch_loss = LossMeter()
meters = (acc_meter, batch_loss)
for lidx, (point_clouds, labels) in enumerate(loader):
optimizer.zero_grad()
data = (point_clouds, labels)
loss = processing(cfg, model, criterion, data, meters)
loss.backward()
optimizer.step()
# nan
if torch.isnan(loss):
print("Train loss is nan.")
exit()
# update lr step
scheduler.step()
# get epoch loss and acc
train_loss = batch_loss.compute()
train_acc = acc_meter.compute()
print('-> Train loss: {} mAcc: {} oAcc: {}'.format(train_loss,
train_acc["class"],
train_acc["overall"]))
# save loss and acc to tensorboard
lr = scheduler.get_last_lr()[0]
log_dict = {
"lr": lr,
"train/loss": train_loss,
"train/mAcc": train_acc["class"],
"train/oAcc": train_acc["overall"]
}
return log_dict
def train(cfg,
model,
loader,
optimizer,
criterion,
scheduler,
publisher="train"):
model.train()
# metrics
acc_meter = MultiAssessmentMeter(num_classes=cfg.dataset.num_classes,
metrics=["class", "overall", "iou"])
batch_loss = LossMeter()
meters = (acc_meter, batch_loss)
for _, data in enumerate(loader):
optimizer.zero_grad()
loss = processing(model, criterion, data, meters, cfg.general.device)
loss.backward()
optimizer.step()
# nan
if torch.isnan(loss):
print("Training loss is nan.")
exit()
scheduler.step()
# get epoch loss and accuracy
epoch_loss = batch_loss.compute()
epoch_acc = acc_meter.compute()
# save loss and acc to tensorboard
lr = scheduler.get_last_lr()[0]
log_dict = {
"lr": lr,
"{}/loss".format(publisher): epoch_loss,
"{}/mAcc".format(publisher): epoch_acc["class"],
"{}/oAcc".format(publisher): epoch_acc["overall"],
"{}/IoU".format(publisher): epoch_acc["iou"]
}
return log_dict
def eval(cfg, model, dataset, criterion, publisher="test"):
model.eval()
loader = DataLoader(dataset,
batch_size=cfg.batch_size,
num_workers=cfg.nworkers,
pin_memory=True)
loader = tqdm(loader, ncols=100, desc=publisher)
acc_meter = MultiAssessmentMeter(num_classes=cfg.num_classes,
metrics=["class", "overall"])
batch_loss = LossMeter()
with torch.no_grad():
for lidx, (point_clouds, labels) in enumerate(loader):
data = (point_clouds, labels)
meters = (acc_meter, batch_loss)
_ = processing(cfg, model, criterion, data, meters)
# get epoch loss and acc
test_loss = batch_loss.compute()
test_acc = acc_meter.compute()
print('-> {} loss: {} mAcc: {} oAcc: {}'.format(publisher, test_loss,
test_acc["class"],
test_acc["overall"]))
def test(cfg, model, dataset, criterion, publisher="test"):
model.eval()
loader = DataLoader(
#Subset(dataset["train"],range(320)),
dataset,
batch_size=cfg.general.batch_size,
num_workers=cfg.loader.nworkers,
pin_memory=True,
)
# loader = tqdm(loader, ncols=100, desc=publisher)
acc_meter = MultiAssessmentMeter(num_classes=dataset.num_classes,
metrics=["class", "overall", "iou"])
batch_loss = LossMeter()
meters = (acc_meter, batch_loss)
with torch.no_grad():
for data in loader:
loss = processing(model, criterion, data, meters,
cfg.general.device)
# get epoch loss and acc
test_loss = batch_loss.compute()
test_acc = acc_meter.compute()
print('-> {} loss: {} mAcc: {} iou: {} oAcc: {}'.format(
publisher, test_loss, test_acc["class"], test_acc["iou"],
test_acc["overall"]))
# save loss and acc to tensorboard
log_dict = {
"{}/loss".format(publisher): test_loss,
"{}/mAcc".format(publisher): test_acc["class"],
"{}/oAcc".format(publisher): test_acc["overall"],
"{}/IoU".format(publisher): test_acc["iou"]
}
return log_dict