asmr/utils.py at main · stevolopolis/asmr · 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
import random
import yaml
import shutil
import numpy as np
import torch
from dataset import *
import torchvision
from omegaconf import OmegaConf


def load_config(config_file):
    configs = yaml.safe_load(open(config_file))
    return configs

def save_src_for_reproduce(configs, out_dir):
    if os.path.exists(os.path.join('outputs', out_dir, 'src')):
        shutil.rmtree(os.path.join('outputs', out_dir, 'src'))
    shutil.copytree('models', os.path.join('outputs', out_dir, 'src', 'models'))
    # dump config to yaml file
    OmegaConf.save(dict(configs), os.path.join('outputs', out_dir, 'src', 'config.yaml'))


def seed_everything(seed: int):
    random.seed(seed)
    np.random.seed(seed) # for random partitioning
    torch.manual_seed(seed)
    torch.cuda.manual_seed(seed)
    torch.backends.cudnn.deterministic = True
    torch.backends.cudnn.benchmark = True


def visualize_activations(activation, H=512, W=768, name='temp_activations', quality=60, downsample=5):
    # make torchvision grid and save as image
    activation = (activation - activation.min(dim=0)[0])/(activation.max(dim=0)[0] - activation.min(dim=0)[0])
    activation = rearrange(activation, '(h w) c -> c 1 h w', h=H, w=W)
    activation = activation.detach().cpu()
    grid = torchvision.utils.make_grid(activation, nrow=16, padding=0)
    img = torchvision.transforms.ToPILImage()(grid)
    img = img.resize((img.size[0]//downsample, img.size[1]//downsample))    # reduce size to save space
    img.save(f"{name}.jpg", format='JPEG', quality=quality)


def ds_profiling(model, data_shape, device=0):
    from deepspeed.accelerator import get_accelerator
    from deepspeed.profiling.flops_profiler import get_model_profile
    # computational cost in number of multiply-accumulates (MACs) per sample
    spatial_size = data_shape[:-1]
    with get_accelerator().device(device):
        flops, macs, params = get_model_profile(model=model, # model
                                        input_shape=data_shape, # input shape to the model. If specified, the model takes a tensor with this shape as the only positional argument.
                                        args=None, # list of positional arguments to the model.
                                        kwargs=None, # dictionary of keyword arguments to the model.
                                        print_profile=True, # prints the model graph with the measured profile attached to each module
                                        detailed=True, # print the detailed profile
                                        module_depth=-1, # depth into the nested modules, with -1 being the inner most modules
                                        top_modules=1, # the number of top modules to print aggregated profile
                                        warm_up=3, # the number of warm-ups before measuring the time of each module
                                        as_string=False, # print raw numbers (e.g. 1000) or as human-readable strings (e.g. 1k)
                                        output_file=None, # path to the output file. If None, the profiler prints to stdout.
                                        ignore_modules=None) # the list of modules to ignore in the profiling
    return flops/np.prod(spatial_size), macs/np.prod(spatial_size), params


def get_dataset(dataset_configs, input_output_configs):
    if  dataset_configs.data_type == "audio":
        dataset = AudioFileDataset(dataset_configs, input_output_configs)
    elif dataset_configs.data_type == "video":
        dataset = VideoFileDataset(dataset_configs, input_output_configs)
    elif dataset_configs.data_type == "image":
        dataset = ImageFileDataset(dataset_configs, input_output_configs)
    elif dataset_configs.data_type == "cameraman":
        dataset = CameraFileDataset(dataset_configs, input_output_configs)
    elif dataset_configs.data_type == "sdf":
        dataset = PointCloud(dataset_configs)
    elif dataset_configs.data_type == "megapixel":
        dataset = BigImageFile(dataset_configs, input_output_configs)

    return dataset


def get_model(model_configs, dataset):
    if model_configs.name == 'ASMR':
        from models.asmr import ASMR
        model = ASMR(
            dim_in=dataset.dim_in,
            dim_out=dataset.dim_out,
            data_size = dataset.get_data_shape()[:-1],
            asmr_configs=model_configs
        )
    elif model_configs.name == 'SIREN':
        from models.siren import Siren
        model = Siren(
            dim_in=dataset.dim_in,
            dim_out=dataset.dim_out,
            siren_configs=model_configs
        )
    elif model_configs.name == 'FFN':
        from models.ffn import FFN
        model = FFN(
            dim_in=dataset.dim_in,
            dim_out=dataset.dim_out,
            ffn_configs=model_configs
        )
    elif model_configs.name == "WIRE":
        from models.wire import Wire
        model = Wire(
           in_features=dataset.dim_in,
           out_features=dataset.dim_out,
           wire_configs=model_configs
        )
    elif model_configs.name == 'KILONERF':
        from models.kilonerf import KiloNerf
        model = KiloNerf(
            in_features=dataset.dim_in,
            out_features=dataset.dim_out,
            data_size = dataset.get_data_shape()[:-1],
            kilo_configs=model_configs
        )
    elif model_configs.name == 'NGP':
        from models.ngp import NGP
        model = NGP(
            dim_in=dataset.dim_in,
            dim_out=dataset.dim_out,
            data_size=dataset.get_data_shape()[:-1],
            ngp_configs=model_configs
        )
    elif model_configs.name == "LOE":
        from models.loe import LoE
        model_configs.NET.img_dim = (dataset.H, dataset.W)
        preset_dim = model_configs.NET.dimensions[-1]
        if dataset.H > dataset.W:
            model_configs.NET.dimensions[-1] = [max(preset_dim), min(preset_dim)] if dataset.dim_in == 2 else [1, 3, 2]
        elif dataset.H < dataset.W:
            model_configs.NET.dimensions[-1] = [min(preset_dim), max(preset_dim)] if dataset.dim_in == 2 else [1, 2, 3]
        print(model_configs.NET.dimensions)

        model = LoE(
           in_features=dataset.dim_in,
           out_features=dataset.dim_out,
           loe_configs=model_configs
        )
    elif model_configs.name == "LOE_SIREN":
        from models.loe import LoE_SIREN
        model_configs.NET.img_dim = (dataset.H, dataset.W)
        preset_dim = model_configs.NET.dimensions[-1]
        if dataset.H > dataset.W:
            model_configs.NET.dimensions[-1] = [max(preset_dim), min(preset_dim)] if dataset.dim_in == 2 else [1, 3, 2]
        elif dataset.H < dataset.W:
            model_configs.NET.dimensions[-1] = [min(preset_dim), max(preset_dim)] if dataset.dim_in == 2 else [1, 2, 3]
        #else:
        #    model_configs.NET.dimensions[-1] = [2, 2] if dataset.dim_in == 2 else [1, 2, 2]
        print(model_configs.NET.dimensions)

        model = LoE_SIREN(
           in_features=dataset.dim_in,
           out_features=dataset.dim_out,
           loe_configs=model_configs
        )
    else:
        raise NotImplementedError(f"Model {model_configs.name} not implemented")

    return model