tijepa/make_mvsa.py at main · nhatkhangcs/tijepa · 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
from eval_on_mvsa import encode_dataset, train_simple_linear_module, MVSA
import os
import sys

TRAIN = "SMALL-A100-448-600-10k-OBS-SCHEDULER-1744373585"
EP = 300

def to_tensor():
    import torchvision.transforms.functional as F
    from torchvision import transforms

    # Tensorize by set tensor_folder=None
    MVSA(
        batch_size=1,
        img_size=448,
        device='cuda:0',
        transform=transforms.Compose(
            [
                transforms.ToTensor()
            ]
        ),
        tensor_folder=None
    )

def encode_all():

    # os.makedirs(f"trains/{TRAIN}/tensors")
    # os.makedirs(f"trains/{TRAIN}/tensors/{EP}-epoch-target")
    # os.makedirs(f"trains/{TRAIN}/tensors/{EP}-epoch-context")

    encode_dataset(
        checkpoint_path=f"trains/{TRAIN}/epoch-{EP}.pt",
        batch_size=200,
        device='cuda:0',
        save_path=f"trains/{TRAIN}/tensors/{EP}-epoch-target",
        crosser_type='target',
        tensor_folder='src/datasets/mvsa-tensor-448-new',
    )
    # encode_dataset(
    #     checkpoint_path=f"trains/{TRAIN}/epoch-{EP}.pt",
    #     batch_size=200,
    #     device='cuda:0',
    #     save_path=f"trains/{TRAIN}/{EP}-epoch-context",
    #     crosser_type='context',
    #     tensor_folder='src/datasets/mvsa-tensor-448',
    # )

def train():
    train_simple_linear_module(
        save_path=f"trains/{TRAIN}/tensors/{EP}-epoch-target",
        hidden_size=1024,
        batch_size=128,
        epochs=50,
        device='cuda:0',
        seed=200, # 100
        lr=1e-3
    )
    # train_simple_linear_module(
    #     save_path=f"trains/{TRAIN}/tensors/{EP}-epoch-context",
    #     hidden_size=768,
    #     batch_size=512,
    #     epochs=50,
    #     device='cuda:0',
    #     seed=100
    # )

if __name__ == "__main__":
    # encode_all()
    train()