def main():
parser = get_parser()
parser.add_argument("--out-stride",
type=int,
default=16,
help="network output stride (default: 8)")
# PASCAL VOC
parser.add_argument(
"--dataset",
type=str,
default="context",
choices=["pascal", "coco", "cityscapes"],
help="dataset name (default: pascal)",
)
parser.add_argument(
"--use-sbd",
action="store_true",
default=True,
help="whether to use SBD dataset (default: True)",
)
parser.add_argument("--base-size",
type=int,
default=513,
help="base image size")
parser.add_argument("--crop-size",
type=int,
default=513,
help="crop image size")
parser.add_argument(
"--loss-type",
type=str,
default="ce",
choices=["ce", "focal"],
help="loss func type (default: ce)",
)
# training hyper params
# PASCAL VOC
parser.add_argument(
"--epochs",
type=int,
default=300,
metavar="N",
help="number of epochs to train (default: auto)",
)
# PASCAL VOC
parser.add_argument(
"--batch-size",
type=int,
default=8,
metavar="N",
help="input batch size for training (default: auto)",
)
# cuda, seed and logging
parser.add_argument(
"--imagenet_pretrained_path",
type=str,
default=
"checkpoint/resnet_backbone_pretrained_imagenet_wo_pascalcontext.pth.tar",
)
parser.add_argument(
"--resume",
type=str,
default=
"checkpoint/deeplab_pascal_context_02_unseen_GMMN_final.pth.tar",
help="put the path to resuming file if needed",
)
parser.add_argument("--checkname", type=str, default="context_eval")
# evaluation option
parser.add_argument("--eval-interval",
type=int,
default=5,
help="evaluation interval (default: 1)")
# keep empty
parser.add_argument("--unseen_classes_idx", type=int, default=[])
# 2 unseen
unseen_names = ["cow", "motorbike"]
# 4 unseen
# unseen_names = ['cow', 'motorbike', 'sofa', 'cat']
# 6 unseen
# unseen_names = ['cow', 'motorbike', 'sofa', 'cat', 'boat', 'fence']
# 8 unseen
# unseen_names = ['cow', 'motorbike', 'sofa', 'cat', 'boat', 'fence', 'bird', 'tvmonitor']
# 10 unseen
# unseen_names = ['cow', 'motorbike', 'sofa', 'cat', 'boat', 'fence', 'bird', 'tvmonitor', 'aeroplane', 'keyboard']
unseen_classes_idx_metric = []
for name in unseen_names:
unseen_classes_idx_metric.append(CLASSES_NAMES.index(name))
### FOR METRIC COMPUTATION IN ORDER TO GET PERFORMANCES FOR TWO SETS
seen_classes_idx_metric = np.arange(60)
seen_classes_idx_metric = np.delete(seen_classes_idx_metric,
unseen_classes_idx_metric).tolist()
parser.add_argument("--seen_classes_idx_metric",
type=int,
default=seen_classes_idx_metric)
parser.add_argument("--unseen_classes_idx_metric",
type=int,
default=unseen_classes_idx_metric)
parser.add_argument("--nonlinear_last_layer",
type=bool,
default=False,
help="non linear prediction")
parser.add_argument("--random_last_layer",
type=bool,
default=False,
help="randomly init last layer")
args = parser.parse_args()
args.cuda = not args.no_cuda and torch.cuda.is_available()
if args.cuda:
try:
args.gpu_ids = [int(s) for s in args.gpu_ids.split(",")]
except ValueError:
raise ValueError(
"Argument --gpu_ids must be a comma-separated list of integers only"
)
args.sync_bn = args.cuda and len(args.gpu_ids) > 1
# default settings for epochs, batch_size and lr
if args.epochs is None:
epoches = {
"coco": 30,
"cityscapes": 200,
"pascal": 50,
}
args.epochs = epoches[args.dataset.lower()]
if args.batch_size is None:
args.batch_size = 4 * len(args.gpu_ids)
if args.test_batch_size is None:
args.test_batch_size = args.batch_size
if args.lr is None:
lrs = {
"coco": 0.1,
"cityscapes": 0.01,
"pascal": 0.007,
}
args.lr = lrs[args.dataset.lower()] / (
4 * len(args.gpu_ids)) * args.batch_size
if args.checkname is None:
args.checkname = "deeplab-resnet"
print(args)
torch.manual_seed(args.seed)
trainer = Trainer(args)
print("Starting Epoch:", trainer.args.start_epoch)
print("Total Epoches:", trainer.args.epochs)
trainer.validation(0, args)
def main():
parser = get_parser()
parser.add_argument(
"--imagenet_pretrained",
type=bool,
default=True,
help="imagenet pretrained backbone",
)
parser.add_argument(
"--out-stride", type=int, default=16, help="network output stride (default: 8)"
)
# PASCAL VOC
parser.add_argument(
"--dataset",
type=str,
default="context",
choices=["pascal", "coco", "cityscapes", "context"],
help="dataset name (default: pascal)",
)
parser.add_argument("--base-size", type=int, default=312, help="base image size")
parser.add_argument("--crop-size", type=int, default=312, help="crop image size")
parser.add_argument(
"--loss-type",
type=str,
default="ce",
choices=["ce", "focal"],
help="loss func type (default: ce)",
)
# training hyper params
# PASCAL VOC
parser.add_argument(
"--epochs",
type=int,
default=200,
metavar="N",
help="number of epochs to train (default: auto)",
)
# PASCAL VOC
parser.add_argument(
"--batch-size",
type=int,
default=10,
metavar="N",
help="input batch size for training (default: auto)",
)
# checking point
parser.add_argument(
"--resume",
type=str,
default=None,
help="put the path to resuming file if needed",
)
parser.add_argument(
"--checkname",
type=str,
default="context_2_unseen",
help="set the checkpoint name",
)
parser.add_argument(
"--imagenet_pretrained_path",
type=str,
default="checkpoint/resnet_backbone_pretrained_imagenet_wo_pascalcontext.pth.tar",
help="set the checkpoint name",
)
# evaluation option
parser.add_argument(
"--eval-interval", type=int, default=10, help="evaluation interval (default: 1)"
)
# 2 unseen
unseen_names = ["cow", "motorbike"]
# 4 unseen
# unseen_names = ['cow', 'motorbike', 'sofa', 'cat']
# 6 unseen
# unseen_names = ['cow', 'motorbike', 'sofa', 'cat', 'boat', 'fence']
# 8 unseen
# unseen_names = ['cow', 'motorbike', 'sofa', 'cat', 'boat', 'fence', 'bird', 'tvmonitor']
# 10 unseen
# unseen_names = ['cow', 'motorbike', 'sofa', 'cat', 'boat', 'fence', 'bird', 'tvmonitor', 'aeroplane', 'keyboard']
unseen_classes_idx = []
for name in unseen_names:
unseen_classes_idx.append(CLASSES_NAMES.index(name))
print(unseen_classes_idx)
# all classes
parser.add_argument("--unseen_classes_idx", type=int, default=unseen_classes_idx)
args = parser.parse_args()
args.cuda = not args.no_cuda and torch.cuda.is_available()
if args.cuda:
try:
args.gpu_ids = [int(s) for s in args.gpu_ids.split(",")]
except ValueError:
raise ValueError(
"Argument --gpu_ids must be a comma-separated list of integers only"
)
args.sync_bn = args.cuda and len(args.gpu_ids) > 1
# default settings for epochs, batch_size and lr
if args.epochs is None:
epoches = {
"coco": 30,
"cityscapes": 200,
"pascal": 50,
"pascal": 150,
}
args.epochs = epoches[args.dataset.lower()]
if args.batch_size is None:
args.batch_size = 4 * len(args.gpu_ids)
if args.test_batch_size is None:
args.test_batch_size = args.batch_size
if args.lr is None:
lrs = {
"coco": 0.1,
"cityscapes": 0.01,
"pascal": 0.007,
}
args.lr = lrs[args.dataset.lower()] / (4 * len(args.gpu_ids)) * args.batch_size
if args.checkname is None:
args.checkname = "deeplab-resnet"
print(args)
torch.manual_seed(args.seed)
trainer = Trainer(args)
print("Starting Epoch:", trainer.args.start_epoch)
print("Total Epoches:", trainer.args.epochs)
for epoch in range(trainer.args.start_epoch, trainer.args.epochs):
trainer.training(epoch)
if not trainer.args.no_val and epoch % args.eval_interval == (
args.eval_interval - 1
):
trainer.validation(epoch)
trainer.writer.close()
def main():
parser = get_parser()
parser.add_argument(
"--imagenet_pretrained",
type=bool,
default=True,
help="imagenet pretrained backbone",
)
parser.add_argument("--out-stride",
type=int,
default=16,
help="network output stride (default: 8)")
# PASCAL VOC
parser.add_argument(
"--dataset",
type=str,
default="pascal",
choices=["pascal", "coco", "cityscapes"],
help="dataset name (default: pascal)",
)
parser.add_argument(
"--use-sbd",
action="store_true",
default=True,
help="whether to use SBD dataset (default: True)",
)
parser.add_argument("--base-size",
type=int,
default=312,
help="base image size")
parser.add_argument("--crop-size",
type=int,
default=312,
help="crop image size")
parser.add_argument(
"--loss-type",
type=str,
default="ce",
choices=["ce", "focal"],
help="loss func type (default: ce)",
)
# training hyper params
# PASCAL VOC
parser.add_argument(
"--epochs",
type=int,
default=200,
metavar="N",
help="number of epochs to train (default: auto)",
)
# PASCAL VOC
parser.add_argument(
"--batch-size",
type=int,
default=16,
metavar="N",
help="input batch size for training (default: auto)",
)
# checking point
parser.add_argument(
"--resume",
type=str,
default=None,
help="put the path to resuming file if needed",
)
parser.add_argument(
"--imagenet_pretrained_path",
type=str,
default=
"checkpoint/resnet_backbone_pretrained_imagenet_wo_pascalvoc.pth.tar",
help="set the checkpoint name",
)
parser.add_argument(
"--checkname",
type=str,
default="pascal_2_unseen",
help="set the checkpoint name",
)
parser.add_argument(
'--config',
type=str,
default='***.yaml',
help='configuration file for train/val',
)
# evaluation option
parser.add_argument("--eval-interval",
type=int,
default=10,
help="evaluation interval (default: 1)")
# only seen classes
# 10 unseen
# parser.add_argument('--unseen_classes_idx', type=int, default=[10, 14, 1, 18, 8, 20, 19, 5, 9, 16])
# 8 unseen
# parser.add_argument('--unseen_classes_idx', type=int, default=[10, 14, 1, 18, 8, 20, 19, 5])
# 6 unseen
# parser.add_argument('--unseen_classes_idx', type=int, default=[10, 14, 1, 18, 8, 20])
# 4 unseen
# parser.add_argument('--unseen_classes_idx', type=int, default=[10, 14, 1, 18])
# 2 unseen
# parser.add_argument("--unseen_classes_idx", type=int, default=[10, 14])
args = parser.parse_args()
args.cuda = not args.no_cuda and torch.cuda.is_available()
if args.cuda:
try:
args.gpu_ids = [int(s) for s in args.gpu_ids.split(",")]
except ValueError:
raise ValueError(
"Argument --gpu_ids must be a comma-separated list of integers only"
)
args.sync_bn = args.cuda and len(args.gpu_ids) > 1
# default settings for epochs, batch_size and lr
if args.epochs is None:
epoches = {
"coco": 30,
"cityscapes": 200,
"pascal": 50,
}
args.epochs = epoches[args.dataset.lower()]
if args.batch_size is None:
args.batch_size = 4 * len(args.gpu_ids)
if args.test_batch_size is None:
args.test_batch_size = args.batch_size
if args.lr is None:
lrs = {
"coco": 0.1,
"cityscapes": 0.01,
"pascal": 0.007,
}
args.lr = lrs[args.dataset.lower()] / (
4 * len(args.gpu_ids)) * args.batch_size
if args.checkname is None:
args.checkname = "deeplab-resnet"
print(args)
torch.manual_seed(args.seed)
trainer = Trainer(args)
print("Starting Epoch:", trainer.args.start_epoch)
print("Total Epoches:", trainer.args.epochs)
for epoch in range(trainer.args.start_epoch, trainer.args.epochs):
trainer.training(epoch)
if not trainer.args.no_val and epoch % args.eval_interval == (
args.eval_interval - 1):
trainer.validation(epoch)
trainer.writer.close()
def main():
parser = get_parser()
parser.add_argument("--out-stride",
type=int,
default=16,
help="network output stride (default: 8)")
# PASCAL VOC
parser.add_argument(
"--dataset",
type=str,
default="context",
choices=["pascal", "coco", "cityscapes"],
help="dataset name (default: pascal)",
)
parser.add_argument(
"--use-sbd",
action="store_true",
default=False,
help="whether to use SBD dataset (default: True)",
)
parser.add_argument("--base-size",
type=int,
default=312,
help="base image size")
parser.add_argument("--crop-size",
type=int,
default=312,
help="crop image size")
parser.add_argument(
"--loss-type",
type=str,
default="ce",
choices=["ce", "focal"],
help="loss func type (default: ce)",
)
# training hyper params
# PASCAL VOC
parser.add_argument(
"--epochs",
type=int,
default=20,
metavar="N",
help="number of epochs to train (default: auto)",
)
# PASCAL VOC
parser.add_argument(
"--batch-size",
type=int,
default=8,
metavar="N",
help="input batch size for training (default: auto)",
)
# checking point
parser.add_argument(
"--imagenet_pretrained_path",
type=str,
default=
"checkpoint/resnet_backbone_pretrained_imagenet_wo_pascalcontext.pth.tar",
)
parser.add_argument(
"--resume",
type=str,
default=
"run/context/context_4_unseen_not_filter_unseen_classes/experiment_0/190_model.pth.tar",
help="put the path to resuming file if needed",
)
parser.add_argument(
"--checkname",
type=str,
default=
"gmmn_context_w2c300_linear_weighted100_hs256_4_unseen_not_filtering",
)
# false if embedding resume
parser.add_argument("--global_avg_pool_bn", type=bool, default=True)
# evaluation option
parser.add_argument("--eval-interval",
type=int,
default=1,
help="evaluation interval (default: 1)")
# keep empty
parser.add_argument("--unseen_classes_idx", type=int, default=[])
# 2 unseen
# unseen_names = ["cow", "motorbike"]
# 4 unseen
unseen_names = ['cow', 'motorbike', 'sofa', 'cat']
# 6 unseen
# unseen_names = ['cow', 'motorbike', 'sofa', 'cat', 'boat', 'fence']
# 8 unseen
# unseen_names = ['cow', 'motorbike', 'sofa', 'cat', 'boat', 'fence', 'bird', 'tvmonitor']
# 10 unseen
# unseen_names = ['cow', 'motorbike', 'sofa', 'cat', 'boat', 'fence', 'bird', 'tvmonitor', 'aeroplane', 'keyboard']
unseen_classes_idx_metric = []
for name in unseen_names:
unseen_classes_idx_metric.append(CLASSES_NAMES.index(name))
### FOR METRIC COMPUTATION IN ORDER TO GET PERFORMANCES FOR TWO SETS
seen_classes_idx_metric = np.arange(33)
seen_classes_idx_metric = np.delete(seen_classes_idx_metric,
unseen_classes_idx_metric).tolist()
parser.add_argument("--seen_classes_idx_metric",
type=int,
default=seen_classes_idx_metric)
parser.add_argument("--unseen_classes_idx_metric",
type=int,
default=unseen_classes_idx_metric)
parser.add_argument("--unseen_weight",
type=int,
default=10,
help="number of output channels")
parser.add_argument("--random_last_layer",
type=bool,
default=True,
help="randomly init last layer")
parser.add_argument(
"--real_seen_features",
type=bool,
default=True,
help="real features for seen classes",
)
parser.add_argument(
"--load_embedding",
type=str,
default="my_w2c",
choices=["attributes", "w2c", "w2c_bg", "my_w2c", "fusion", None],
)
parser.add_argument("--w2c_size", type=int, default=600)
### GENERATOR ARGS
parser.add_argument("--noise_dim", type=int, default=600)
parser.add_argument("--embed_dim", type=int, default=600)
parser.add_argument("--hidden_size", type=int, default=256)
parser.add_argument("--feature_dim", type=int, default=256)
parser.add_argument("--lr_generator", type=float, default=0.0002)
parser.add_argument("--batch_size_generator", type=int, default=128)
parser.add_argument("--saved_validation_images", type=int, default=10)
parser.add_argument(
"--filter_unseen_classes",
type=bool,
default=False,
help="filter unseen classes",
)
parser.add_argument(
'--config',
type=str,
default='configs/context_finetune.yaml',
help='configuration file for train/val',
)
args = parser.parse_args()
args.cuda = not args.no_cuda and torch.cuda.is_available()
if args.cuda:
try:
args.gpu_ids = [int(s) for s in args.gpu_ids.split(",")]
except ValueError:
raise ValueError(
"Argument --gpu_ids must be a comma-separated list of integers only"
)
args.sync_bn = args.cuda and len(args.gpu_ids) > 1
# default settings for epochs, batch_size and lr
if args.epochs is None:
epoches = {
"coco": 30,
"cityscapes": 200,
"pascal": 50,
}
args.epochs = epoches[args.dataset.lower()]
if args.batch_size is None:
args.batch_size = 4 * len(args.gpu_ids)
if args.test_batch_size is None:
args.test_batch_size = args.batch_size
if args.lr is None:
lrs = {
"coco": 0.1,
"cityscapes": 0.01,
"pascal": 0.007,
}
args.lr = lrs[args.dataset.lower()] / (
4 * len(args.gpu_ids)) * args.batch_size
if args.checkname is None:
args.checkname = "deeplab-resnet"
print(args)
torch.manual_seed(args.seed)
trainer = Trainer(args)
print("Starting Epoch:", trainer.args.start_epoch)
print("Total Epoches:", trainer.args.epochs)
# trainer.validation(trainer.args.start_epoch, args)
for epoch in range(trainer.args.start_epoch, trainer.args.epochs):
trainer.training(epoch, args)
if not trainer.args.no_val and epoch % args.eval_interval == (
args.eval_interval - 1):
trainer.validation(epoch, args)
trainer.writer.close()
def main():
parser = get_parser()
parser.add_argument("--out-stride",
type=int,
default=16,
help="network output stride (default: 8)")
# PASCAL VOC
parser.add_argument(
"--dataset",
type=str,
default="pascal",
choices=["pascal", "coco", "cityscapes"],
help="dataset name (default: pascal)",
)
parser.add_argument(
"--use-sbd",
action="store_true",
default=False,
help="whether to use SBD dataset (default: True)",
)
parser.add_argument("--base-size",
type=int,
default=513,
help="base image size")
parser.add_argument("--crop-size",
type=int,
default=513,
help="crop image size")
parser.add_argument(
"--loss-type",
type=str,
default="ce",
choices=["ce", "focal"],
help="loss func type (default: ce)",
)
# training hyper params
# PASCAL VOC
parser.add_argument(
"--epochs",
type=int,
default=300,
metavar="N",
help="number of epochs to train (default: auto)",
)
# PASCAL VOC
parser.add_argument(
"--batch-size",
type=int,
default=8,
metavar="N",
help="input batch size for training (default: auto)",
)
# cuda, seed and logging
# checking point
parser.add_argument(
"--imagenet_pretrained_path",
type=str,
default=
"checkpoint/resnet_backbone_pretrained_imagenet_wo_pascalvoc.pth.tar",
)
# checking point
parser.add_argument(
"--resume",
type=str,
default=
"run/pascal/gmmn_pascal_w2c300_linear_weighted100_hs256_2_unseen/experiment_0/26_model.pth.tar",
help="put the path to resuming file if needed",
)
parser.add_argument("--checkname", type=str, default="pascal_eval")
# evaluation option
parser.add_argument("--eval-interval",
type=int,
default=5,
help="evaluation interval (default: 1)")
### FOR IMAGE SELECTION IN ORDER TO TAKE OFF IMAGE WITH UNSEEN CLASSES FOR TRAINING AND VALIDATION
# keep empty
parser.add_argument("--unseen_classes_idx", type=int, default=[])
### FOR METRIC COMPUTATION IN ORDER TO GET PERFORMANCES FOR TWO SETS
seen_classes_idx_metric = np.arange(20)
# 2 unseen
unseen_classes_idx_metric = [10, 14]
# 4 unseen
# unseen_classes_idx_metric = [10, 14, 1, 18]
# 6 unseen
# unseen_classes_idx_metric = [10, 14, 1, 18, 8, 20]
# 8 unseen
# unseen_classes_idx_metric = [10, 14, 1, 18, 8, 20, 19, 5]
# 10 unseen
# unseen_classes_idx_metric = [10, 14, 1, 18, 8, 20, 19, 5, 9, 16]
# 5 unseen
unseen_classes_idx_metric = [15, 16, 17, 18, 19]
seen_classes_idx_metric = np.delete(seen_classes_idx_metric,
unseen_classes_idx_metric).tolist()
parser.add_argument("--seen_classes_idx_metric",
type=int,
default=seen_classes_idx_metric)
parser.add_argument("--unseen_classes_idx_metric",
type=int,
default=unseen_classes_idx_metric)
parser.add_argument("--nonlinear_last_layer",
type=bool,
default=False,
help="non linear prediction")
parser.add_argument("--random_last_layer",
type=bool,
default=False,
help="randomly init last layer")
args = parser.parse_args()
args.cuda = not args.no_cuda and torch.cuda.is_available()
if args.cuda:
try:
args.gpu_ids = [int(s) for s in args.gpu_ids.split(",")]
except ValueError:
raise ValueError(
"Argument --gpu_ids must be a comma-separated list of integers only"
)
args.sync_bn = args.cuda and len(args.gpu_ids) > 1
# default settings for epochs, batch_size and lr
if args.epochs is None:
epoches = {
"coco": 30,
"cityscapes": 200,
"pascal": 50,
}
args.epochs = epoches[args.dataset.lower()]
if args.batch_size is None:
args.batch_size = 4 * len(args.gpu_ids)
if args.test_batch_size is None:
args.test_batch_size = args.batch_size
if args.lr is None:
lrs = {
"coco": 0.1,
"cityscapes": 0.01,
"pascal": 0.007,
}
args.lr = lrs[args.dataset.lower()] / (
4 * len(args.gpu_ids)) * args.batch_size
if args.checkname is None:
args.checkname = "deeplab-resnet"
print(args)
torch.manual_seed(args.seed)
trainer = Trainer(args)
print("Starting Epoch:", trainer.args.start_epoch)
print("Total Epoches:", trainer.args.epochs)
trainer.validation(0, args)
