brain-dissection/arguments.py at main · Gabesarch/brain-dissection · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
import argparse
import numpy as np
import ipdb
st = ipdb.set_trace
parser = argparse.ArgumentParser()

parser.add_argument("--set_name", type=str, default="test00", help="name of experiment")
parser.add_argument("--seed", type=int, default=42, help="Random seed")
parser.add_argument("--mode", type=str, help="mode to run, see main.py")
parser.add_argument("--split", type=str, default="val", help="dataloader split")


############## VIP #########
# parser.add_argument("--arch", type=str, default="dino", help="which arch to use for training?")
parser.add_argument("--pretraining", type=str, default="dino", help="which pretraining to load weights from?")
parser.add_argument("--distributed", action="store_true", default=False, help="distributed training")
parser.add_argument('--lr', default=2e-5, type=float)
parser.add_argument('--clip_max_norm', default=0.1, type=float,
                    help='gradient clipping max norm')

parser.add_argument('--weight_decay', default=1e-2, type=float)
parser.add_argument('--epochs', default=100, type=int)
parser.add_argument('--lr_drop', default=4000, type=int)
parser.add_argument('--readout_sparse_weight', default=0.02, type=float)
parser.add_argument("--sparsity_on_feature_weights", action="store_true", default=False, help="put sparsity of feature weights for what/where")
parser.add_argument("--val_freq", type=int, default=2500, help="how often to run validation")
parser.add_argument("--arch", type=str, default="cnn_alt", help="which arch to use for training?")
parser.add_argument("--patience", type=int, default=19, help="patience for early stopping")
parser.add_argument("--early_stopping", type=int, default=10, help="Stop training if correlation has not improved upon best in X epochs. Patience must also be defined.")


parser.add_argument('--total_voxel_size', default=None, type=int)

parser.add_argument("--max_iters", type=int, default=50000, help="maximum iterations to train")
parser.add_argument("--load_model", action="store_true", default=False, help="Load an existing checkpoint")
parser.add_argument("--load_model_path", type=str, default="", help="Path to existing checkpoint")
parser.add_argument("--load_strict_false", action="store_true", default=False, help="do not load strict checkpoint")
parser.add_argument("--lr_scheduler_from_scratch", action="store_true", default=False, help="do not load LR scheduler from checkpoint if True")
parser.add_argument("--optimizer_from_scratch", action="store_true", default=False, help="do not load optimizer from checkpoint if True")
parser.add_argument("--start_one", action="store_true", default=False, help="start from iteration 0")
parser.add_argument('--batch_size', default=24, type=int, help="batch size for model training")
parser.add_argument('--num_workers', default=4, type=int)
parser.add_argument('--save_freq_epoch', default=5, type=int, help="How often every X epochs to save model")
parser.add_argument('--keep_latest', default=10, type=int, help="number of checkpoints to keep at one time")
parser.add_argument('--max_validation_iters', default=None, type=int, help="maximum validation iters")
parser.add_argument("--checkpoint_path", type=str, default="./checkpoints", help="Path for saving checkpoints")
parser.add_argument("--attention_threshold", type=float, default=None, help="""We visualize masks
    obtained by thresholding the self-attention maps to keep xx% of the mass.""")
parser.add_argument("--images_path", type=str, default="./data/images", help="Path for saving checkpoints")

parser.add_argument("--activation_threshold", type=float, default=0.99, help="""We get IOUs
    by thresholding the activation maps""")

# parser.add_argument('--patch_size', default=16, type=int, help="patch size for vit in pixels")

parser.add_argument("--log_freq", type=int, default=250, help="how often to log to tensorboard in iterations")
# parser.add_argument("--lr_scheduler_freq", type=int, default=500, help="how often to step LR scheduler in iterations")
parser.add_argument("--run_validation", action="store_true", default=False, help="run validation every val_freq iters")
parser.add_argument("--save_freq", type=int, default=500, help="how often to save a checkpoint")

parser.add_argument("--image_size", type=int, default=224, help="")
parser.add_argument("--image_size_eval", type=int, default=56, help="")

parser.add_argument('--readout_sparse_weight_spatial', default=0.0, type=float)
parser.add_argument('--readout_sparse_weight_feature', default=0.0, type=float)

parser.add_argument("--coco_images_path", type=str, default="", help="Path to coco images")
parser.add_argument("--coco_annotations_path", type=str, default="", help="Path to coco annotations")

parser.add_argument("--subjects_repeat_path", type=str, default="", help="Path to subject image indices")
parser.add_argument("--brain_data_dir", type=str, default="", help="Path to nsd brain data")
parser.add_argument("--roi_dir", type=str, default="", help="Path to roi info")
parser.add_argument("--noise_ceiling_dir", type=str, default="", help="Path to noise ceiling info")

parser.add_argument("--nc_threshold", type=float, default=0.1, help="noise ceiling threshold for removal")
parser.add_argument("--debug", action="store_true", default=False, help="debugging mode?")

parser.add_argument("--analyze_depth", action="store_true", default=False, help="add depth analysis to dissection")

parser.add_argument("--rois", type=str, nargs='+', default=['FFA-1', 'FFA-2'], help="which rois to use?")
parser.add_argument("--coco_ids_path", type=str, default="", help="Path to coco ids to filter")

parser.add_argument("--max_activation_plots", type=int, default=None, help="plot every X units")
parser.add_argument("--subsample_activations", type=int, default=None, help="subsample every X activations")

parser.add_argument("--default_args", type=str, default=None, help="set default args")

parser.add_argument("--subjects", type=int, nargs='+', default=[1,2,3,4,5,6,7,8], help="which subjects to use?")

parser.add_argument("--topk", type=int, default=10, help="")

parser.add_argument("--subsample_images", type=int, default=None, help="subsample every X images")

parser.add_argument("--group", type=str, default="default", help="group name")

parser.add_argument("--min_test_corr", type=float, default=None, help="minimum correlation to keep unit")

parser.add_argument("--tmp_dir", type=str, default="", help="Path to nsd tmp folder")
parser.add_argument("--max_images", type=int, default=None, help="max images for debugging")
parser.add_argument("--eval_subject", type=int, default=1, help="which subject to use for eval?")


parser.add_argument("--topk_units_from_corr", type=int, default=None, help="take topk units from test correlations")

# broden dataset
parser.add_argument("--categories", type=int, nargs='+', default=["object", "part","scene","texture","color"], help="categories to fetch for dataset")
parser.add_argument("--data_directory", type=str, default="./dataset/broden1_224", help="Path to images")


parser.add_argument("--images_path_alt", type=str, default=None, help="Path to alternate images")

parser.add_argument("--save_dissection_samples", action="store_true", default=False, help="save dissection data?")
parser.add_argument("--load_dissection_samples", action="store_true", default=False, help="save dissection data?")


parser.add_argument("--filter_images_by_responses", action="store_true", default=False, help="filter images by top preicted responses for each ROI?")
parser.add_argument("--topk_filter", type=int, default=1000, help="topk to keep for filter_images_by_responses")

parser.add_argument("--wandb_directory", type=str, default='./wandb', help="Path to wandb metadata")


parser.add_argument("--gqa_path", type=str, default='./gqa', help="Path to wandb metadata")
parser.add_argument("--shared_depth_maps", action="store_true", default=False, help="shared depth maps across subjects?")
parser.add_argument("--use_spatial_mask", action="store_true", default=False, help="use spatial mask in eval?")

parser.add_argument("--reduced_eval_memory", action="store_true", default=False, help="use spatial mask in eval?")

parser.add_argument("--xtc_checkpoint_paths", type=str, default='./checkpoints', help="Path to wandb metadata")
parser.add_argument("--mode2", default=None, type=str, help="mode to run, see main.py")

parser.add_argument("--tmp_dir_load", default=None, type=str, help="mode to run, see main.py")

parser.add_argument("--eval_object", default=None, type=str, help="mode to run, see main.py")

parser.add_argument("--plot_for_figure", action="store_true", default=False, help="generate plot for figure in XTC eval")

parser.add_argument("--figure_unit_start", type=int, default=0, help="")
parser.add_argument("--figure_unit_end", type=int, default=20, help="")

parser.add_argument("--randomly_shuffle_images", action="store_true", default=False, help="randomly shuffle images in response dataloader?")
parser.add_argument("--randomly_shuffle_labels", action="store_true", default=False, help="randomly shuffle labels in places dataloader?")

parser.add_argument("--load_model_paths_unitvisual", type=str, nargs='+', default=[], help="which model paths to use for unit visualization?")

args = parser.parse_args()

if args.debug and args.mode in ["convnet_nsd_response_optimized", "convnet_get_corrs"]:
    args.subjects = [1]
    args.run_validation = True

if args.debug and args.mode in ["convnet_nsd_eval_baudissect", "convnet_places_eval_baudissect", "convnet_broden_eval_baudissect", "convnet_ade20k_eval_baudissect", "convnet_gqa_eval_baudissect", "convnet_xtc_eval_baudissect"]:
    args.image_size_eval = 56
    args.topk = 20
    args.max_images = 50
    args.save_dissection_samples = False
    args.load_dissection_samples = True