if self.args['GENERAL']['debug']:
model_utils.check_occupation(complet_inputs['complet_input'],
batch_complet.dense())
complet_output = self.complet_head(batch_complet)
torch.cuda.empty_cache()
return seg_output, complet_output
classifier = J3SC_Net()
if use_cuda:
classifier = classifier.cuda()
classifier = classifier.eval()
scn.checkpoint_restore(classifier, model_path, use_cuda)
print('#classifer parameters %d' %
sum([x.nelement() for x in classifier.parameters()]))
dataset = importlib.import_module('kitti_dataset')
input_data = kitti_dataset.get_dataset(config, split=args.dataset)
data_loader = torch.utils.data.DataLoader(
input_data,
batch_size=1,
collate_fn=seg_head.Merge,
num_workers=config['TRAIN']['train_workers'],
pin_memory=True,
shuffle=False,
drop_last=False,
worker_init_fn=lambda x: np.random.seed(x + int(time.time())))
num_sample = len(data_loader)
scn.BatchNormReLU(m)).add(
scn.OutputLayer(data.dimension))
self.linear = nn.Linear(m, 20)
def forward(self, x):
x = self.sparseModel(x)
x = self.linear(x)
return x
unet = Model()
if use_cuda:
unet = unet.cuda()
training_epochs = 2048
training_epoch = scn.checkpoint_restore(unet, exp_name, 'unet', use_cuda)
optimizer = optim.Adam(unet.parameters())
print('#classifer parameters', sum([x.nelement() for x in unet.parameters()]))
for epoch in range(training_epoch, training_epochs + 1):
unet.train()
stats = {}
scn.forward_pass_multiplyAdd_count = 0
scn.forward_pass_hidden_states = 0
start = time.time()
train_loss = 0
for i, batch in enumerate(data.train_data_loader):
optimizer.zero_grad()
if use_cuda:
batch['x'][1] = batch['x'][1].cuda()
batch['y'] = batch['y'].cuda()
requires_grad=True)
def forward(self, x):
seg_output, _ = self.seg_head(x)
return seg_output
classifier = J3SC_Net(config)
print(classifier)
if use_cuda:
classifier = classifier.cuda()
classifier = classifier.eval()
training_epoch = scn.checkpoint_restore(classifier, model_path, use_cuda)
print('#classifer parameters %d' %
sum([x.nelement() for x in classifier.parameters()]))
'''Load Dataset'''
config_file = os.path.join('opt/SemanticPOSS.yaml')
poss_config = yaml.safe_load(open(config_file, 'r'))
scan = laserscan.SemLaserScan(nclasses=11,
sem_color_dict=poss_config['color_map'])
sequences = poss_config['split']['valid']
label_to_names = {
0: 'people',
1: 'rider',
2: 'car',
3: 'trunk',
4: 'plants',