def get_pytorch_train_loader(self):
print("in get_pytorch_train_loader fucntion")
pipe_train = trainPipeline(self.data_path, self.batch_size,
self.num_thread, self.crop, self.rali_cpu)
pipe_train.build()
train_loader = RALIClassificationIterator(pipe_train)
return train_loader
def main():
if len(sys.argv) < 4:
print('Please pass image_folder cpu/gpu batch_size ')
exit(0)
if (sys.argv[2] == "cpu"):
_rali_cpu = True
else:
_rali_cpu = False
bs = int(sys.argv[3])
nt = 1
di = 0
crop_size = 224
image_path = sys.argv[1]
pipe = HybridTrainPipe(batch_size=bs,
num_threads=nt,
device_id=di,
data_dir=image_path,
crop=crop_size,
rali_cpu=_rali_cpu)
pipe.build()
imageIterator = RALIClassificationIterator(pipe)
num_classes = len(next(os.walk(image_path))[1])
print("num_classes:: ", num_classes)
net = ToyNet(num_classes)
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(), lr=0.0005, momentum=0.9)
# Training loop
for epoch in range(10): # loop over the dataset multiple times
print("epoch:: ", epoch)
running_loss = 0.0
for i, (image_batch, labels) in enumerate(imageIterator, 0):
sys.stdout.write("\r Mini-batch " + str(i))
# print("Images",image_batch)
# print("Labels",labels)
optimizer.zero_grad()
outputs = net(image_batch)
loss = criterion(outputs, labels)
loss.backward()
optimizer.step()
# print statistics
running_loss += loss.item()
print_interval = 10
if i % print_interval == (print_interval - 1):
print('[%d, %5d] loss: %.3f' %
(epoch + 1, i + 1, running_loss / print_interval))
running_loss = 0.0
print('Finished Training')
def main():
if len(sys.argv) < 4:
print(
'Please pass image_folder cpu/gpu batch_size classification/detection'
)
exit(0)
if (sys.argv[2] == "cpu"):
_rali_cpu = True
else:
_rali_cpu = False
bs = int(sys.argv[3])
if (sys.argv[4] == "classification"):
_rali_type = True
else:
_rali_type = False
nt = 1
di = 0
crop_size = 224
image_path = sys.argv[1]
pipe = HybridTrainPipe(batch_size=bs,
num_threads=nt,
device_id=di,
data_dir=image_path,
crop=crop_size,
rali_cpu=_rali_cpu,
rali_type=_rali_type)
pipe.build()
imageIterator = RALIClassificationIterator(pipe)
num_classes = len(next(os.walk(image_path))[1])
print("num_classes:: ", num_classes)
# Training loop
for epoch in range(1): # loop over the dataset multiple times
print("epoch:: ", epoch)
if _rali_type:
for i, (image_batch, labels) in enumerate(imageIterator,
0): # Classification
sys.stdout.write("\r Mini-batch " + str(i))
print("Images", image_batch)
print("Labels", labels)
else:
for i, (image_batch, bboxes,
labels) in enumerate(imageIterator, 0): # Detection
sys.stdout.write("\r Mini-batch " + str(i))
print("Images", image_batch)
print("Bboxes", bboxes)
print("Labels", labels)
print('Finished Training')
def get_pytorch_val_loader(self):
pipe_val = valPipeline(self.data_path, self.batch_size,
self.num_thread, self.crop, self.rali_cpu)
pipe_val.build()
val_loader = RALIClassificationIterator(pipe_val)
return val_loader