mirrors
/
deep-person-reid
mirror of https://github.com/KaiyangZhou/deep-person-reid.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity

update args.py

pull/119/head
KaiyangZhou 2019-02-03 14:03:22 +00:00
parent bac2ce5662
commit 7076f1a750
1 changed files with 17 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
args.py
View File

@ -23,6 +23,8 @@ def argument_parser():
help='split index (note: 0-based)') help='split index (note: 0-based)')
parser.add_argument('--train-sampler', type=str, default='RandomSampler', parser.add_argument('--train-sampler', type=str, default='RandomSampler',
help='sampler for trainloader') help='sampler for trainloader')
parser.add_argument('--augdata-re', action='store_true',
help='use random erasing for data augmentation')
# ************************************************************ # ************************************************************
# Video datasets # Video datasets
@ -96,6 +98,13 @@ def argument_parser():
parser.add_argument('--open-layers', type=str, nargs='+', default=['classifier'], parser.add_argument('--open-layers', type=str, nargs='+', default=['classifier'],
help='open specified layers for training while keeping others frozen') help='open specified layers for training while keeping others frozen')
parser.add_argument('--staged-lr', action='store_true',
help='set different lr to different layers')
parser.add_argument('--new-layers', type=str, nargs='+', default=['classifier'],
help='newly added layers with default lr')
parser.add_argument('--base-lr-mult', type=float, default=0.1,
help='learning rate multiplier for base layers')
# ************************************************************ # ************************************************************
# Cross entropy loss-specific setting # Cross entropy loss-specific setting
# ************************************************************ # ************************************************************
@ -175,7 +184,8 @@ def image_dataset_kwargs(parsed_args):
'num_instances': parsed_args.num_instances, 'num_instances': parsed_args.num_instances,
'cuhk03_labeled': parsed_args.cuhk03_labeled, 'cuhk03_labeled': parsed_args.cuhk03_labeled,
'cuhk03_classic_split': parsed_args.cuhk03_classic_split, 'cuhk03_classic_split': parsed_args.cuhk03_classic_split,
'market1501_500k': parsed_args.market1501_500k 'market1501_500k': parsed_args.market1501_500k,
'augdata_re': parsed_args.augdata_re,
} }
@ -197,7 +207,8 @@ def video_dataset_kwargs(parsed_args):
'train_sampler': parsed_args.train_sampler, 'train_sampler': parsed_args.train_sampler,
'num_instances': parsed_args.num_instances, 'num_instances': parsed_args.num_instances,
'seq_len': parsed_args.seq_len, 'seq_len': parsed_args.seq_len,
'sample_method': parsed_args.sample_method 'sample_method': parsed_args.sample_method,
'augdata_re': parsed_args.augdata_re,
} }
@ -215,5 +226,8 @@ def optimizer_kwargs(parsed_args):
'sgd_nesterov': parsed_args.sgd_nesterov, 'sgd_nesterov': parsed_args.sgd_nesterov,
'rmsprop_alpha': parsed_args.rmsprop_alpha, 'rmsprop_alpha': parsed_args.rmsprop_alpha,
'adam_beta1': parsed_args.adam_beta1, 'adam_beta1': parsed_args.adam_beta1,
'adam_beta2': parsed_args.adam_beta2 'adam_beta2': parsed_args.adam_beta2,
'staged_lr': parsed_args.staged_lr,
'new_layers': parsed_args.new_layers,
'base_lr_mult': parsed_args.base_lr_mult,
} }