format="png",
cmap="hot")
# fig = plt.imshow(np.transpose(npimg, (1, 2, 0)))
# plt.axis('off')
# fig.axes.get_xaxis().set_visible(False)
# fig.axes.get_yaxis().set_visible(False)
# plt.savefig(img_path, bbox_inches='tight', pad_inches = 0)
# plt.imshow(np.transpose(npimg, (1, 2, 0)))
# plt.savefig(img_path)
net = CAN()
#net.load_state_dict(torch.load('best_val_tensorboard.wts'))
net.load_state_dict(checkpoint["model_state"])
net.to(device)
net.eval()
i = 0
save_data = True
running_metrics_val = runningScore(2)
with torch.no_grad():
for data in trainloader:
imgs, labels = data
imgs, labels = imgs.to(device), labels.to(device)
if save_data:
save_img(imgs, 'images/' + str(i) + '.png', True)
np.transpose(npimg, (1, 2, 0)),
format="png",
cmap="hot")
# fig = plt.imshow(np.transpose(npimg, (1, 2, 0)))
# plt.axis('off')
# fig.axes.get_xaxis().set_visible(False)
# fig.axes.get_yaxis().set_visible(False)
# plt.savefig(img_path, bbox_inches='tight', pad_inches = 0)
# plt.imshow(np.transpose(npimg, (1, 2, 0)))
# plt.savefig(img_path)
net = CAN()
net.load_state_dict(torch.load('best_val_tensorboard.wts'))
net.to(device)
net.eval()
i = 0
save_data = False
running_metrics_val = runningScore(2)
with torch.no_grad():
for data in trainloader:
imgs, labels = data
imgs, labels = imgs.to(device), labels.to(device)
if save_data:
save_img(imgs, 'images/' + str(i) + '.png', True)
pretrained_state_dict['features.23.weight'] = pretrained_state_dict.pop(
'features.24.weight')
pretrained_state_dict['features.23.bias'] = pretrained_state_dict.pop(
'features.24.bias')
pretrained_state_dict['features.25.weight'] = pretrained_state_dict.pop(
'features.26.weight')
pretrained_state_dict['features.25.bias'] = pretrained_state_dict.pop(
'features.26.bias')
pretrained_state_dict['features.27.weight'] = pretrained_state_dict.pop(
'features.28.weight')
pretrained_state_dict['features.27.bias'] = pretrained_state_dict.pop(
'features.28.bias')
#net.load_state_dict(torch.load('trained.wts'))
net.load_state_dict(pretrained_state_dict, strict=False)
net.to(device)
net.train()
augmentations = Compose([RandomRotate(5), RandomHorizontallyFlip()])
train_dataset = tusimpleLoader('/home/tejus/Downloads/train_set/',
split="train",
augmentations=augmentations)
trainloader = torch.utils.data.DataLoader(train_dataset,
batch_size=TRAIN_BATCH,
shuffle=True,
num_workers=TRAIN_BATCH,
pin_memory=True)
val_dataset = tusimpleLoader('/home/tejus/Downloads/train_set/',
split="val",
# loss_fn = nn.BCEWithLogitsLoss()
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer, 'min', patience = 2, verbose = True, min_lr = 0.000001)
loss_fn = nn.CrossEntropyLoss()
net.to(device)
if not resume_training:
pretrained_state_dict = models.vgg16(pretrained=True).state_dict()
pretrained_state_dict = {k:v for k, v in pretrained_state_dict.items() if 'classifier' not in k}
#rename parameters. ordering changes because maxpool layers aren't present in network.
pretrained_state_dict['features.23.weight'] = pretrained_state_dict.pop('features.24.weight')
pretrained_state_dict['features.23.bias'] = pretrained_state_dict.pop('features.24.bias')
pretrained_state_dict['features.25.weight'] = pretrained_state_dict.pop('features.26.weight')
pretrained_state_dict['features.25.bias'] = pretrained_state_dict.pop('features.26.bias')
pretrained_state_dict['features.27.weight'] = pretrained_state_dict.pop('features.28.weight')
pretrained_state_dict['features.27.bias'] = pretrained_state_dict.pop('features.28.bias')
net.load_state_dict(pretrained_state_dict, strict=False)
start_iter = 0
else:
checkpoint = torch.load(checkpoint_dir)
net.load_state_dict(checkpoint["model_state"])
optimizer.load_state_dict(checkpoint["optimizer_state"])
scheduler.load_state_dict(checkpoint["scheduler_state"])
start_iter = checkpoint["epoch"]
logger.info(
"Loaded checkpoint '{}' (epoch {})".format(
checkpoint_dir, checkpoint["epoch"]
)
)
net.train()