def main():
opt = OptInit().get_args()
logging.info('===> Creating dataloader ...')
train_dataset = GeoData.S3DIS(opt.data_dir,
opt.area,
True,
pre_transform=T.NormalizeScale())
train_loader = DenseDataLoader(train_dataset,
batch_size=opt.batch_size,
shuffle=True,
num_workers=4)
test_dataset = GeoData.S3DIS(opt.data_dir,
opt.area,
train=False,
pre_transform=T.NormalizeScale())
test_loader = DenseDataLoader(test_dataset,
batch_size=opt.batch_size,
shuffle=False,
num_workers=0)
opt.n_classes = train_loader.dataset.num_classes
logging.info('===> Loading the network ...')
model = DenseDeepGCN(opt).to(opt.device)
if opt.multi_gpus:
model = DataParallel(DenseDeepGCN(opt)).to(opt.device)
logging.info('===> loading pre-trained ...')
model, opt.best_value, opt.epoch = load_pretrained_models(
model, opt.pretrained_model, opt.phase)
logging.info(model)
logging.info('===> Init the optimizer ...')
criterion = torch.nn.CrossEntropyLoss().to(opt.device)
optimizer = torch.optim.Adam(model.parameters(), lr=opt.lr)
scheduler = torch.optim.lr_scheduler.StepLR(optimizer, opt.lr_adjust_freq,
opt.lr_decay_rate)
optimizer, scheduler, opt.lr = load_pretrained_optimizer(
opt.pretrained_model, optimizer, scheduler, opt.lr)
logging.info('===> Init Metric ...')
opt.losses = AverageMeter()
opt.test_value = 0.
logging.info('===> start training ...')
for _ in range(opt.epoch, opt.total_epochs):
opt.epoch += 1
logging.info('Epoch:{}'.format(opt.epoch))
train(model, train_loader, optimizer, scheduler, criterion, opt)
if opt.epoch % opt.eval_freq == 0 and opt.eval_freq != -1:
test(model, test_loader, opt)
scheduler.step()
logging.info('Saving the final model.Finish!')
def main():
opt = OptInit().initialize()
opt.batch_size = 1
os.environ["CUDA_VISIBLE_DEVICES"] = opt.gpuNum
print('===> Creating dataloader...')
# def __init__(self,
# root,
# is_train=True,
# is_validation=False,
# is_test=False,
# num_channel=5,
# pre_transform=None,
# pre_filter=None)
test_dataset = BigredDataset(root=opt.test_path,
is_train=False,
is_validation=False,
is_test=True,
num_channel=5,
pre_transform=T.NormalizeScale())
test_loader = DenseDataLoader(test_dataset,
batch_size=opt.batch_size,
shuffle=False,
num_workers=32)
opt.n_classes = 2
print('len(test_loader):', len(test_loader))
print('phase: ', opt.phase)
print('batch_size: ', opt.batch_size)
print('use_cpu: ', opt.use_cpu)
print('gpuNum: ', opt.gpuNum)
print('multi_gpus: ', opt.multi_gpus)
print('test_path: ', opt.test_path)
print('in_channels: ', opt.in_channels)
print('device: ', opt.device)
print('===> Loading the network ...')
model = DenseDeepGCN(opt).to(opt.device)
load_package = torch.load(opt.pretrained_model)
model, opt.best_value, opt.epoch = load_pretrained_models(
model, opt.pretrained_model, opt.phase)
pdb.set_trace()
for item in load_package.keys():
if (item != 'optimizer_state_dict' and item != 'state_dict'
and item != 'scheduler_state_dict'):
print(str(item), load_package[item])
print('===> Start Evaluation ...')
test(model, test_loader, opt)
def main():
opt = OptInit().initialize()
opt.printer.info('===> Creating dataloader ...')
train_dataset = GeoData.S3DIS(opt.train_path,
5,
True,
pre_transform=T.NormalizeScale())
train_loader = DenseDataLoader(train_dataset,
batch_size=opt.batch_size,
shuffle=True,
num_workers=4)
opt.n_classes = train_loader.dataset.num_classes
opt.printer.info('===> Loading the network ...')
model = DenseDeepGCN(opt).to(opt.device)
if opt.multi_gpus:
model = DataParallel(DenseDeepGCN(opt)).to(opt.device)
opt.printer.info('===> loading pre-trained ...')
model, opt.best_value, opt.epoch = load_pretrained_models(
model, opt.pretrained_model, opt.phase)
opt.printer.info('===> Init the optimizer ...')
criterion = torch.nn.CrossEntropyLoss().to(opt.device)
if opt.optim.lower() == 'adam':
optimizer = torch.optim.Adam(model.parameters(), lr=opt.lr)
elif opt.optim.lower() == 'radam':
optimizer = optim.RAdam(model.parameters(), lr=opt.lr)
else:
raise NotImplementedError('opt.optim is not supported')
scheduler = torch.optim.lr_scheduler.StepLR(optimizer, opt.lr_adjust_freq,
opt.lr_decay_rate)
optimizer, scheduler, opt.lr = load_pretrained_optimizer(
opt.pretrained_model, optimizer, scheduler, opt.lr)
opt.printer.info('===> Init Metric ...')
opt.losses = AverageMeter()
# opt.test_metric = miou
# opt.test_values = AverageMeter()
opt.test_value = 0.
opt.printer.info('===> start training ...')
for _ in range(opt.epoch, opt.total_epochs):
opt.epoch += 1
train(model, train_loader, optimizer, scheduler, criterion, opt)
# test_value = test(model, test_loader, test_metric, opt)
scheduler.step()
opt.printer.info('Saving the final model.Finish!')
def main():
opt = OptInit().get_args()
logging.info('===> Creating dataloader...')
test_dataset = GeoData.S3DIS(opt.data_dir,
opt.area,
train=False,
pre_transform=T.NormalizeScale())
test_loader = DenseDataLoader(test_dataset,
batch_size=opt.batch_size,
shuffle=False,
num_workers=0)
opt.n_classes = test_loader.dataset.num_classes
if opt.no_clutter:
opt.n_classes -= 1
logging.info('===> Loading the network ...')
model = DenseDeepGCN(opt).to(opt.device)
model, opt.best_value, opt.epoch = load_pretrained_models(
model, opt.pretrained_model, opt.phase)
logging.info('===> Start Evaluation ...')
test(model, test_loader, opt)
cur_shape_iou_tot += I/U
cur_shape_iou_cnt += 1.
if cur_shape_iou_cnt > 0:
cur_shape_miou = cur_shape_iou_tot / cur_shape_iou_cnt
shape_iou_tot += cur_shape_miou
shape_iou_cnt += 1.
shape_mIoU = shape_iou_tot / shape_iou_cnt
part_iou = np.divide(part_intersect[1:], part_union[1:])
mean_part_iou = np.mean(part_iou)
opt.printer.info("===> Category {}-{}, Part mIOU is{:.4f} \t ".format(
opt.category_no, opt.category, mean_part_iou))
if __name__ == '__main__':
opt = OptInit().initialize()
opt.printer.info('===> Creating dataloader ...')
test_dataset = PartNet(opt.data_dir, opt.dataset, opt.category, opt.level, 'val')
test_loader = DenseDataLoader(test_dataset, batch_size=1, shuffle=True, num_workers=1)
opt.n_classes = test_loader.dataset.num_classes
opt.printer.info('===> Loading the network ...')
model = DenseDeepGCN(opt).to(opt.device)
opt.printer.info('===> loading pre-trained ...')
model, opt.best_value, opt.epoch = load_pretrained_models(model, opt.pretrained_model, opt.phase)
test(model, test_loader, opt)
def main():
opt = OptInit().get_args()
logging.info('===> Creating dataloader ...')
train_dataset = GeoData.S3DIS(opt.data_dir,
opt.area,
True,
pre_transform=T.NormalizeScale())
train_loader = DenseDataLoader(train_dataset,
batch_size=opt.batch_size,
shuffle=True,
num_workers=4)
test_dataset = GeoData.S3DIS(opt.data_dir,
opt.area,
train=False,
pre_transform=T.NormalizeScale())
test_loader = DenseDataLoader(test_dataset,
batch_size=opt.batch_size,
shuffle=False,
num_workers=0)
opt.n_classes = train_loader.dataset.num_classes
logging.info('===> Loading the network ...')
model = DenseDeepGCN(opt).to(opt.device)
if opt.multi_gpus:
model = DataParallel(DenseDeepGCN(opt)).to(opt.device)
logging.info('===> loading pre-trained ...')
model, opt.best_value, opt.epoch = load_pretrained_models(
model, opt.pretrained_model, opt.phase)
logging.info(model)
logging.info('===> Init the optimizer ...')
criterion = torch.nn.CrossEntropyLoss().to(opt.device)
optimizer = torch.optim.Adam(model.parameters(), lr=opt.lr)
scheduler = torch.optim.lr_scheduler.StepLR(optimizer, opt.lr_adjust_freq,
opt.lr_decay_rate)
optimizer, scheduler, opt.lr = load_pretrained_optimizer(
opt.pretrained_model, optimizer, scheduler, opt.lr)
logging.info('===> Init Metric ...')
opt.losses = AverageMeter()
opt.test_value = 0.
logging.info('===> start training ...')
for _ in range(opt.epoch, opt.total_epochs):
opt.epoch += 1
logging.info('Epoch:{}'.format(opt.epoch))
train(model, train_loader, optimizer, criterion, opt)
if opt.epoch % opt.eval_freq == 0 and opt.eval_freq != -1:
test(model, test_loader, opt)
scheduler.step()
# ------------------ save checkpoints
# min or max. based on the metrics
is_best = (opt.test_value < opt.best_value)
opt.best_value = max(opt.test_value, opt.best_value)
model_cpu = {k: v.cpu() for k, v in model.state_dict().items()}
save_checkpoint(
{
'epoch': opt.epoch,
'state_dict': model_cpu,
'optimizer_state_dict': optimizer.state_dict(),
'scheduler_state_dict': scheduler.state_dict(),
'best_value': opt.best_value,
}, is_best, opt.ckpt_dir, opt.exp_name)
# ------------------ tensorboard log
info = {
'loss': opt.losses.avg,
'test_value': opt.test_value,
'lr': scheduler.get_lr()[0]
}
opt.writer.add_scalars('epoch', info, opt.iter)
logging.info('Saving the final model.Finish!')
def main():
NUM_POINT = 20000
opt = OptInit().initialize()
opt.num_worker = 32
os.environ["CUDA_VISIBLE_DEVICES"] = opt.gpuNum
opt.printer.info('===> Creating dataloader ...')
train_dataset = BigredDataset(root = opt.train_path,
is_train=True,
is_validation=False,
is_test=False,
num_channel=opt.num_channel,
pre_transform=T.NormalizeScale()
)
train_loader = DenseDataLoader(train_dataset, batch_size=opt.batch_size, shuffle=False, num_workers=opt.num_worker)
validation_dataset = BigredDataset(root = opt.train_path,
is_train=False,
is_validation=True,
is_test=False,
num_channel=opt.num_channel,
pre_transform=T.NormalizeScale()
)
validation_loader = DenseDataLoader(validation_dataset, batch_size=opt.batch_size, shuffle=True, num_workers=opt.num_worker)
opt.printer.info('===> computing Labelweight ...')
labelweights = np.zeros(2)
labelweights, _ = np.histogram(train_dataset.data.y.numpy(), range(3))
labelweights = labelweights.astype(np.float32)
labelweights = labelweights / np.sum(labelweights)
labelweights = np.power(np.amax(labelweights) / labelweights, 1 / 3.0)
weights = torch.Tensor(labelweights).cuda()
print("labelweights", weights)
opt.n_classes = train_loader.dataset.num_classes
opt.printer.info('===> Loading the network ...')
opt.best_value = 0
print("GPU:",opt.device)
model = DenseDeepGCN(opt).to(opt.device)
if opt.multi_gpus:
model = DataParallel(DenseDeepGCN(opt)).to(device=opt.device)
opt.printer.info('===> loading pre-trained ...')
# model, opt.best_value, opt.epoch = load_pretrained_models(model, opt.pretrained_model, opt.phase)
opt.printer.info('===> Init the optimizer ...')
criterion = torch.nn.CrossEntropyLoss(weight = weights).to(opt.device)
# criterion_test = torch.nn.CrossEntropyLoss(weight = weights)
if opt.optim.lower() == 'adam':
optimizer = torch.optim.Adam(model.parameters(), lr=opt.lr)
elif opt.optim.lower() == 'radam':
optimizer = optim.RAdam(model.parameters(), lr=opt.lr)
else:
raise NotImplementedError('opt.optim is not supported')
scheduler = torch.optim.lr_scheduler.StepLR(optimizer, opt.lr_adjust_freq, opt.lr_decay_rate)
# optimizer, scheduler, opt.lr = load_pretrained_optimizer(opt.pretrained_model, optimizer, scheduler, opt.lr)
opt.printer.info('===> Init Metric ...')
opt.losses = AverageMeter()
# opt.test_metric = miou
opt.test_values = AverageMeter()
opt.test_value = 0.
opt.printer.info('===> start training ...')
writer = SummaryWriter()
writer_test = SummaryWriter()
counter_test = 0
counter_play = 0
start_epoch = 0
mean_miou = AverageMeter()
mean_loss = AverageMeter()
mean_acc = AverageMeter()
best_value = 0
for epoch in range(start_epoch, opt.total_epochs):
opt.epoch += 1
model.train()
total_seen_class = [0 for _ in range(opt.n_classes)]
total_correct_class = [0 for _ in range(opt.n_classes)]
total_iou_deno_class = [0 for _ in range(opt.n_classes)]
ave_mIoU = 0
total_correct = 0
total_seen = 0
loss_sum = 0
mean_miou.reset()
mean_loss.reset()
mean_acc.reset()
for i, data in tqdm(enumerate(train_loader), total=len(train_loader), smoothing=0.9):
# if i % 50 == 0:
opt.iter += 1
if not opt.multi_gpus:
data = data.to(opt.device)
target = data.y
batch_label2 = target.cpu().data.numpy()
inputs = torch.cat((data.pos.transpose(2, 1).unsqueeze(3), data.x.transpose(2, 1).unsqueeze(3)), 1)
inputs = inputs[:, :opt.num_channel, :, :]
gt = data.y.to(opt.device)
# ------------------ zero, output, loss
optimizer.zero_grad()
out = model(inputs)
loss = criterion(out, gt)
#pdb.set_trace()
# ------------------ optimization
loss.backward()
optimizer.step()
seg_pred= out.transpose(2,1)
pred_val = seg_pred.contiguous().cpu().data.numpy()
seg_pred = seg_pred.contiguous().view(-1, opt.n_classes)
#pdb.set_trace()
pred_val = np.argmax(pred_val, 2)
batch_label = target.view(-1, 1)[:, 0].cpu().data.numpy()
target = target.view(-1, 1)[:, 0]
pred_choice = seg_pred.cpu().data.max(1)[1].numpy()
correct = np.sum(pred_choice == batch_label)
total_correct += correct
total_seen += (opt.batch_size *NUM_POINT)
loss_sum += loss.item()
current_seen_class = [0 for _ in range(opt.n_classes)]
current_correct_class = [0 for _ in range(opt.n_classes)]
current_iou_deno_class = [0 for _ in range(opt.n_classes)]
#pdb.set_trace()
for l in range(opt.n_classes):
#pdb.set_trace()
total_seen_class[l] += np.sum((batch_label2 == l))
total_correct_class[l] += np.sum((pred_val == l) & (batch_label2 == l))
total_iou_deno_class[l] += np.sum(((pred_val == l) | (batch_label2 == l)))
current_seen_class[l] = np.sum((batch_label2 == l))
current_correct_class[l] = np.sum((pred_val == l) & (batch_label2 == l))
current_iou_deno_class[l] = np.sum(((pred_val == l) | (batch_label2 == l)))
#pdb.set_trace()
writer.add_scalar('training_loss', loss.item(), counter_play)
writer.add_scalar('training_accuracy', correct / float(opt.batch_size * NUM_POINT), counter_play)
m_iou = np.mean(np.array(current_correct_class) / (np.array(current_iou_deno_class, dtype=np.float) + 1e-6))
writer.add_scalar('training_mIoU', m_iou, counter_play)
ave_mIoU = np.mean(np.array(total_correct_class) / (np.array(total_iou_deno_class, dtype=np.float) + 1e-6))
# print("training_loss:",loss.item())
# print('training_accuracy:',correct / float(opt.batch_size * NUM_POINT))
# print('training_mIoU:',m_iou)
mean_miou.update(m_iou)
mean_loss.update(loss.item())
mean_acc.update(correct / float(opt.batch_size * NUM_POINT))
counter_play = counter_play + 1
train_mIoU = mean_miou.avg
train_macc = mean_acc.avg
train_mloss = mean_loss.avg
print('Epoch: %d, Training point avg class IoU: %f' % (epoch,train_mIoU))
print('Epoch: %d, Training mean loss: %f' %(epoch, train_mloss))
print('Epoch: %d, Training accuracy: %f' %(epoch, train_macc))
mean_miou.reset()
mean_loss.reset()
mean_acc.reset()
print('validation_loader')
model.eval()
with torch.no_grad():
for i, data in tqdm(enumerate(validation_loader), total=len(validation_loader), smoothing=0.9):
# if i % 50 ==0:
if not opt.multi_gpus:
data = data.to(opt.device)
target = data.y
batch_label2 = target.cpu().data.numpy()
inputs = torch.cat((data.pos.transpose(2, 1).unsqueeze(3), data.x.transpose(2, 1).unsqueeze(3)), 1)
inputs = inputs[:, :opt.num_channel, :, :]
gt = data.y.to(opt.device)
out = model(inputs)
loss = criterion(out, gt)
#pdb.set_trace()
seg_pred = out.transpose(2, 1)
pred_val = seg_pred.contiguous().cpu().data.numpy()
seg_pred = seg_pred.contiguous().view(-1, opt.n_classes)
pred_val = np.argmax(pred_val, 2)
batch_label = target.view(-1, 1)[:, 0].cpu().data.numpy()
target = target.view(-1, 1)[:, 0]
pred_choice = seg_pred.cpu().data.max(1)[1].numpy()
correct = np.sum(pred_choice == batch_label)
current_seen_class = [0 for _ in range(opt.n_classes)]
current_correct_class = [0 for _ in range(opt.n_classes)]
current_iou_deno_class = [0 for _ in range(opt.n_classes)]
for l in range(opt.n_classes):
current_seen_class[l] = np.sum((batch_label2 == l))
current_correct_class[l] = np.sum((pred_val == l) & (batch_label2 == l))
current_iou_deno_class[l] = np.sum(((pred_val == l) | (batch_label2 == l)))
m_iou = np.mean(
np.array(current_correct_class) / (np.array(current_iou_deno_class, dtype=np.float) + 1e-6))
mean_miou.update(m_iou)
mean_loss.update(loss.item())
mean_acc.update(correct / float(opt.batch_size * NUM_POINT))
validation_mIoU = mean_miou.avg
validation_macc = mean_acc.avg
validation_mloss = mean_loss.avg
writer.add_scalar('validation_loss', validation_mloss, epoch)
print('Epoch: %d, validation mean loss: %f' %(epoch, validation_mloss))
writer.add_scalar('validation_accuracy', validation_macc, epoch)
print('Epoch: %d, validation accuracy: %f' %(epoch, validation_macc))
writer.add_scalar('validation_mIoU', validation_mIoU, epoch)
print('Epoch: %d, validation point avg class IoU: %f' % (epoch,validation_mIoU))
model_cpu = {k: v.cpu() for k, v in model.state_dict().items()}
package ={
'epoch': opt.epoch,
'state_dict': model_cpu,
'optimizer_state_dict': optimizer.state_dict(),
'scheduler_state_dict': scheduler.state_dict(),
'train_miou':train_mIoU,
'train_accuracy':train_macc,
'train_loss':train_mloss,
'validation_mIoU':validation_mIoU,
'validation_macc':validation_macc,
'validation_mloss':validation_mloss,
'num_channel':opt.num_channel,
'gpuNum': opt.gpuNum,
'time':time.ctime()
}
torch.save(package,'saves/val_miou_%f_val_acc_%f_%d.pth' % (validation_mIoU, validation_macc, epoch))
is_best = (best_value < validation_mIoU)
print('Is Best? ',is_best)
if (best_value < validation_mIoU):
best_value = validation_mIoU
torch.save(package,'saves/best_model.pth')
print('Best IoU: %f' % (best_value))
scheduler.step()
opt.printer.info('Saving the final model.Finish!')