def simsiam_default(debug=False):
config = DottedDict()
config.fs_ckpt = "model_{}_epoch_{:0>6}.ckpt"
config.mean_std = [[0.485, 0.456, 0.406], [0.229, 0.224, 0.225]]
config.dataset = "imagenet"
config.backbone = "resnet50"
config.batch_size = 512
config.num_epochs = 100
config.img_size = 224
config.projector_args = {
"hidden_dim": 2048,
"out_dim": 2048,
"n_hidden_layers": 1
}
config.predictor_args = {
"hidden_dim": config.projector_args["out_dim"] // 4, # see Appendix B.
"in_dim": config.projector_args["out_dim"],
"out_dim": config.projector_args["out_dim"]
}
config.optimizer = "sgd"
config.base_lr = 0.05
config.optimizer_args = {
"lr": config.base_lr * (config.batch_size / 256),
"weight_decay": 0.0001, # used always
"momentum": 0.9
}
config.scheduler = "cosine_decay"
config.scheduler_args = {
"T_max": config.num_epochs,
"eta_min": 0,
}
config.debug = False
config.num_workers = 8
config.device = 'cuda:0' if torch.cuda.is_available() else 'cpu'
config.resume = False
#
# Frequencies (epochs)
#
config.freq_knn = 1
#
# debug settings
if config.debug:
config.batch_size = 2
config.num_epochs = 5 # train only one epoch
config.num_workers = 1
return config
def linear_default(debug=False):
config = DottedDict()
config.optimizer = "sgd"
config.device = 'cuda:1' if torch.cuda.is_available() else 'cpu'
config.batch_size = 256
config.base_lr = 0.5
config.num_workers = 8
config.num_epochs = 100
config.optimizer_args = {
"lr": config.base_lr,
"weight_decay": 5e-4, # adapted, paper says 0
"momentum": 0.9
}
config.scheduler = "cosine_decay"
config.scheduler_args = {"T_max": config.num_epochs, "eta_min": 0}
return config