LayerInbetween/main4_ctrlpoint_train.py at main · MarkMoHR/LayerInbetween · 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
import json
import os
import six
from configs.example_configs import test_data_base, test_img_id, do_inv, gen_time, example_info_map, gpu_id
from configs.model_configs import model_config_map

os.environ['CUDA_VISIBLE_DEVICES'] = gpu_id
os.environ["KMP_WARNINGS"] = "0"

import model4_ctrlpoint_train as sketch_ctrlpoint_model
from utils4_ctrlpoint_train import load_dataset, load_real_dataset


def trainer(model_params):
    print('Hyperparams:')
    for key, val in six.iteritems(model_params.values()):
        print('%s = %s' % (key, str(val)))
    print('-' * 100)

    datasets = load_dataset(model_params)

    sub_log_root = os.path.join(model_params.log_root, model_params.workspace)
    sub_log_img_root = os.path.join(model_params.log_img_root, model_params.workspace)
    sub_snapshot_root = os.path.join(model_params.snapshot_root, model_params.workspace)
    os.makedirs(sub_log_root, exist_ok=True)
    os.makedirs(sub_log_img_root, exist_ok=True)
    os.makedirs(sub_snapshot_root, exist_ok=True)

    train_set = datasets[0]
    val_set = datasets[1]
    train_model_params = datasets[2]
    val_model_params = datasets[3]

    # Write config file to json file.
    with open(os.path.join(sub_snapshot_root, 'model_config.json'), 'w') as f:
        json.dump(train_model_params.values(), f, indent=True)

    model = sketch_ctrlpoint_model.FullModel(model_params, train_set, val_set,
                                            sub_log_root, sub_snapshot_root, sub_log_img_root)
    model.train()
    model.evaluate()


def tester(model_params, mode):
    print('Hyperparams:')
    for key, val in six.iteritems(model_params.values()):
        print('%s = %s' % (key, str(val)))
    print('-' * 100)

    # TODO: change parameters here
    occluded_only = False
    stroke_fixing = False
    if stroke_fixing: assert occluded_only

    datasets = load_dataset(model_params, test_only=True, stroke_fixing=stroke_fixing)

    train_set = datasets[0]
    val_set = datasets[1]

    sub_snapshot_root = os.path.join(model_params.snapshot_root, model_params.workspace)

    model = sketch_ctrlpoint_model.FullModel(model_params, train_set, val_set,
                                            None, sub_snapshot_root, None)
    if mode == 'inference':
        sub_inference_root = os.path.join(model_params.inference_root, model_params.workspace)
        sub_inference_root += '---' + model_params.transform_local_model_name
        os.makedirs(sub_inference_root, exist_ok=True)
        model.inference(sub_inference_root, show_data='selected')  # ['selected', 'all', 'occluded']
    elif mode == 'inference_full':
        sub_inference_root = os.path.join(model_params.inference_full_root, model_params.workspace)
        sub_inference_root += '---' + model_params.transform_local_model_name
        sub_inference_root += '-[c_min=' + str(model_params.window_size_min_comp) + ']'
        if model_params.use_optical_flow:
            sub_inference_root += '-[optical]'
        os.makedirs(sub_inference_root, exist_ok=True)
        model.inference_full(sub_inference_root, show_data='all', occluded_only=occluded_only)  # ['selected', 'all']
    else:
        model.evaluate(load_trained_weights=True, occluded_only=occluded_only)


def tester_real(model_params, mode):
    print('Hyperparams:')
    for key, val in six.iteritems(model_params.values()):
        print('%s = %s' % (key, str(val)))
    print('-' * 100)

    data_base = test_data_base
    datasets = load_real_dataset(model_params, data_base=data_base, generation_time=gen_time)
    val_set = datasets[0]

    sub_snapshot_root = os.path.join(model_params.snapshot_root, model_params.workspace)

    model = sketch_ctrlpoint_model.FullModel(model_params, None, val_set,
                                            None, sub_snapshot_root, None)
    if mode == 'inference_full_real':
        inference_full_real_root = model_params.inference_full_real_root
        if gen_time > 0:
            inference_full_real_root += '-Gen%d' % gen_time
        os.makedirs(inference_full_real_root, exist_ok=True)
        model.inference_full_real(inference_full_real_root)
    else:
        raise Exception('Unknown mode:', mode)


def tester_real_inv(model_params, mode):
    print('Hyperparams:')
    for key, val in six.iteritems(model_params.values()):
        print('%s = %s' % (key, str(val)))
    print('-' * 100)

    data_base = test_data_base
    data_base_extra = "outputs/stroke_correspondence_results"
    if gen_time > 0:
        data_base_extra += '-Gen%d' % gen_time
    data_base = os.path.join(data_base, '[0inv]')
    datasets = load_real_dataset(model_params, data_base=data_base, data_base_extra=data_base_extra,
                                 generation_time=gen_time)
    val_set = datasets[0]

    sub_snapshot_root = os.path.join(model_params.snapshot_root, model_params.workspace)

    model = sketch_ctrlpoint_model.FullModel(model_params, None, val_set,
                                            None, sub_snapshot_root, None)

    inference_full_real_root = model_params.inference_full_real_root
    if gen_time > 0:
        inference_full_real_root += '-Gen%d' % gen_time
    inference_full_real_root = os.path.join(inference_full_real_root, '[0inv]')
    os.makedirs(inference_full_real_root, exist_ok=True)
    model.inference_full_real(inference_full_real_root, do_inv=True)


if __name__ == '__main__':
    mode = 'inference_full_real'  # ['train', 'test', 'inference', 'inference_full', 'inference_full_real']
    if do_inv:
        mode += '_inv'

    model_params = sketch_ctrlpoint_model.get_default_hparams()

    if 'real' in mode:
        # Add params for real inference
        model_params.add_hparam('use_optical_flow', example_info_map[str(test_img_id)]['use_optical_flow'])
        if not do_inv:
            model_params.add_hparam('use_target_layer', example_info_map[str(test_img_id)]['use_target_layer'])
            model_params.add_hparam('use_target_layer_mask', example_info_map[str(test_img_id)]['use_target_layer_mask'])
        else:
            model_params.add_hparam('use_target_layer', False)
            model_params.add_hparam('use_target_layer_mask', 'stroke')
        model_params.add_hparam('target_layer_method', example_info_map[str(test_img_id)]['target_layer_method'])
    else:
        model_params.add_hparam('use_optical_flow', True)

    model_params.add_hparam('raster_size', model_config_map['raster_size'])
    model_params.add_hparam('window_size_scaling_ref_comp', model_config_map['window_size_scaling_ref_comp'])
    model_params.add_hparam('window_size_min_comp', model_config_map['window_size_min_comp'])
    model_params.add_hparam('window_size_scaling_ref_comp_local', model_config_map['window_size_scaling_ref_comp_local'])
    model_params.add_hparam('window_size_min_comp_local', model_config_map['window_size_min_comp_local'])

    model_params.add_hparam('window_size_scaling_ref', model_config_map['window_size_scaling_ref_ctrl'])
    model_params.add_hparam('window_size_min', model_config_map['window_size_min_ctrl'])

    if mode == 'train':
        trainer(model_params)
    elif mode == 'test' or mode == 'inference' or mode == 'inference_full':
        tester(model_params, mode)
    elif mode == 'inference_full_real':
        tester_real(model_params, mode)
    elif mode == 'inference_full_real_inv':
        tester_real_inv(model_params, mode)
    else:
        raise Exception('Unknown mode:', mode)