player_classification/main.py at main · IPCV/player_classification · 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
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
import copy
import json
import warnings
from argparse import ArgumentParser
from pathlib import Path

import torch
import wandb
from addict import Dict

from clustering import cluster
from test.backbone import test_backbone
from test.cluster import test_cluster
from test.cnn import test_cnn
from test.transformer import test_transformer
from train.backbone import train_backbone
from train.transformer import train_transformer_single_match, train_transformer_multiple_matches, find_lr
from util.misc import init_wandb, load_snv3_sequences, load_snv3_splits, load_snv2_splits, init_weights_dir
from util.misc import load_conf
from util.misc import set_random_seed


def train_and_test(conf, args, device, split=None):
    args = copy.deepcopy(args)

    set_random_seed(conf.optimization.random_seed)
    init_wandb(conf, args.architecture, mode=args.log_mode)

    if not args.only_test:
        if split is not None:
            wandb.run.name += f'_{split.name}'
        init_weights_dir(args)

        if args.architecture == 'transformer':
            train_transformer_single_match(split, conf, args, device)
        else:
            train_backbone(conf, args, device)

        args.checkpoint = args.weights_dir.joinpath('model.pth')

    if args.architecture == 'transformer':
        test_transformer(split.test, conf, args, device)
    else:
        test_backbone(conf, args, device)
    wandb.finish()


if __name__ == '__main__':
    parser = ArgumentParser(description='TransKit training')
    inputs_group = parser.add_mutually_exclusive_group()
    inputs_group.add_argument('-m', '--matches',
                              help='Path for a file containing the SoccerNetV2 training matches list (default: None)',
                              default=None, type=lambda p: Path(p))
    inputs_group.add_argument('-s', '--splits',
                              help='SoccerNetV2 splits filepath (default: None)',
                              default=None, type=lambda p: Path(p))
    inputs_group.add_argument('--sequences',
                              help='JSON file containing SoccerNetV3 sequences ordered by half matches (default: None)',
                              default=None, type=lambda p: Path(p))
    parser.add_argument('-a', '--architecture', required=False,
                        help='Architecture component to train/test (default: transformer)',
                        default='transformer', choices=['backbone', 'cnn', 'transformer'])
    parser.add_argument('-c', '--conf', required=False,
                        help='JSON configuration filepath (default: config/transformer.json)',
                        default="config/transformer.json", type=lambda p: Path(p))
    parser.add_argument('-d', '--dataset_path', required=False,
                        help='Path for SoccerNet dataset (default: data/soccernet)',
                        default="data/soccernet", type=lambda p: Path(p))
    weights_group = parser.add_mutually_exclusive_group()
    weights_group.add_argument('-b', '--backbone_weights', required=False,
                               help='Backbone weights filepath (default: None)',
                               default=None, type=lambda p: Path(p))
    weights_group.add_argument('-t', '--transformer_weights', required=False,
                               help='Transformer weights filepath (default: None)',
                               default=None, type=lambda p: Path(p))
    parser.add_argument('--only_test', required=False,
                        help='Only perform test (default: False)',
                        action='store_true')
    parser.add_argument('--lr_finder', required=False,
                        help='Learning rate finder (default: False)',
                        action='store_true')
    parser.add_argument('--save_predictions', required=False,
                        help='Filename for saving predictions (default: None)',
                        default=None, type=str)
    parser.add_argument('--cluster', required=False,
                        help='Cluster the players from SoccerNet matches (default: False)',
                        action='store_true')
    parser.add_argument('-k', '--kit_clusters_csv', required=False,
                        help='Kit uniform clusters CSV filepath for the SoccerNet matches (default: None)',
                        default=None, type=lambda p: Path(p))
    parser.add_argument('--weights_dir', required=False,
                        help='Weights directory path (default: weights)',
                        default='weights', type=lambda p: Path(p))
    parser.add_argument('--log_mode', required=False,
                        help='Log mode [online, offline, disabled] (default: online)',
                        default='online', type=str)
    parser.add_argument('--log_freq', required=False,
                        help='Results log frequency (default: 20)',
                        default=20, type=int)
    parser.add_argument('--checkpoint', required=False,
                        help='Training checkpoint filepath (default: None)',
                        default=None, type=lambda p: Path(p))
    parser.add_argument('--num_loading_threads', required=False,
                        help='Number of data loading threads (default: 12)',
                        default=12, type=int)
    parser.add_argument('--num_workers', required=False,
                        help='Number of workers for PyTorch dataloaders (default: 2)',
                        default=2, type=int)
    args = parser.parse_args()

    args.is_soccernet_v3 = args.sequences is not None
    args.individual_matches = True
    if args.is_soccernet_v3:
        args.sequences = load_snv3_sequences(args.dataset_path, args.sequences)
    elif args.matches is not None:
        with args.matches.open() as f:
            args.matches = [args.dataset_path.joinpath(m) for m in f.read().splitlines()]
    else:
        with args.splits.open() as json_file:
            args.splits = Dict(json.load(json_file))
            for s in ['train', 'valid', 'test']:
                args.splits[s] = [args.dataset_path.joinpath(m) for m in args.splits[s]]
        args.individual_matches = False

    conf = load_conf(args.conf, args.architecture, args.backbone_weights, args.transformer_weights)

    device = torch.device("cuda")

    set_random_seed(conf.optimization.random_seed)

    if args.cluster:
        if not args.only_test:
            cluster(conf, args, device)
        test_cluster(conf, args, device)
    elif args.architecture == 'backbone':
        train_and_test(conf, args, device)
    elif args.architecture == 'cnn':
        if not args.only_test:
            warnings.warn("Warning: Training CNNs on SoccerNetV2 is not supported for CNNs")

        if args.is_soccernet_v3:
            raise Exception('Training/Testing CNNs on SoccerNetV3 is not implemented for CNNs')

        wandb.init(mode='disabled')
        for match_path in args.matches:
            test_cnn(match_path, conf, args, device)
    elif args.architecture == 'transformer':
        if args.individual_matches:
            if args.is_soccernet_v3:
                splits = load_snv3_splits(args.sequences, conf.optimization.split_train_ratio)
                for split in splits:
                    train_and_test(conf, args, device, split)
            else:
                for match_path in args.matches:
                    try:
                        split = load_snv2_splits(match_path, conf)
                        train_and_test(conf, args, device, split)
                    except:
                        print(match_path)
                        wandb.finish()
        elif not args.lr_finder:
            if not args.only_test:
                init_wandb(conf, args.architecture, mode=args.log_mode)
                init_weights_dir(args)

                train_transformer_multiple_matches(conf, args, device)

            args.checkpoint = args.weights_dir.joinpath('model.pth')
            for match_path in args.splits.test:
                test_transformer(match_path, conf, args, device)
        else:
            init_wandb(conf, args.architecture, mode=args.log_mode)
            init_weights_dir(args)

            find_lr(conf, args, device)