Deep learning models for change detection of remote sensing images

Hi, your project change_detection.pytorch(commit id: 0a86d51b31276d9c413798ab3fb332889f02d8aa) requires "albumentations==1.0.3" in its dependency. After analyzing the source code, we found that the following versions of albumentations can also be suitable, i.e., albumentations 1.0.0, 1.0.1, 1.0.2, since all functions that you directly (8 APIs: albumentations.core.transforms_interface.BasicTransform.init, albumentations.augmentations.geometric.resize.Resize.init, albumentations.core.composition.Compose.init, albumentations.pytorch.transforms.ToTensorV2.init, albumentations.augmentations.crops.functional.random_crop, albumentations.core.transforms_interface.DualTransform.init, albumentations.augmentations.crops.transforms.RandomCrop.init, albumentations.augmentations.transforms.Normalize.init) or indirectly (propagate to 11 albumentations's internal APIs and 0 outsider APIs) used from the package have not been changed in these versions, thus not affecting your usage.

Therefore, we believe that it is quite safe to loose your dependency on albumentations from "albumentations==1.0.3" to "albumentations>=1.0.0,<=1.0.3". This will improve the applicability of change_detection.pytorch and reduce the possibility of any further dependency conflict with other projects.

May I pull a request to further loosen the dependency on albumentations?

By the way, could you please tell us whether such an automatic tool for dependency analysis may be potentially helpful for maintaining dependencies easier during your development?

您好，我在运行local_test.py文件时出现了错误，而我一直解决不了，错误如下： RuntimeError: Expected 4-dimensional input for 4-dimensional weight [64, 3, 7, 7], but got 3-dimensional input of size [3, 256, 256] instead 我想知道[6,3,7,7]代表的是什么？ 这个错误是发生在valed部分，在执行epoch1时train可以正常读取图片并运行，但到valed就报错了，希望能获得您的建议。

This piece of code :

for x in np.arange(0.6, 0.9, 0.1):
    print('Eval with TH:', x)
    metrics = [
        cdp.utils.metrics.Fscore(activation='argmax2d', threshold=x),
        cdp.utils.metrics.Precision(activation='argmax2d', threshold=x),
        cdp.utils.metrics.Recall(activation='argmax2d', threshold=x),
    ]

    valid_epoch = cdp.utils.train.ValidEpoch(
        model,
        loss=loss,
        metrics=metrics,
        device=DEVICE,
        verbose=True,
    )

    valid_logs = valid_epoch.run(valid_loader)
    print(valid_logs)

Give me the following result:

Eval with TH: 0.6
valid: 100%|█████████████████████████████████████████████████████████████| 505/505 [01:12<00:00,  6.98it/s, cross_entropy_loss - 0.08708, fscore - 0.8799, precision - 0.8946, recall - 0.8789]
{'cross_entropy_loss': 0.0870812193864016, 'fscore': 0.8798528309538921, 'precision': 0.8946225793644936, 'recall': 0.8789094516579565}

Eval with TH: 0.7
valid: 100%|█████████████████████████████████████████████████████████████| 505/505 [01:12<00:00,  6.99it/s, cross_entropy_loss - 0.08708, fscore - 0.8799, precision - 0.8946, recall - 0.8789]
{'cross_entropy_loss': 0.08708121913835626, 'fscore': 0.8798528309538921, 'precision': 0.8946225793644936, 'recall': 0.8789094516579565}

Eval with TH: 0.7999999999999999
valid: 100%|█████████████████████████████████████████████████████████████| 505/505 [01:11<00:00,  7.02it/s, cross_entropy_loss - 0.08708, fscore - 0.8799, precision - 0.8946, recall - 0.8789]
{'cross_entropy_loss': 0.08708121978843793, 'fscore': 0.8798528309538921, 'precision': 0.8946225793644936, 'recall': 0.8789094516579565}

Hi @likyoo ,

I study with yoru repo for my project.I have been added to new features to your repo.I'll share it with you when I'm done.

But I have a significant question;

How can I load weigths after training operation?

Dear @likyoo thanks for your open source project. I have trained models and saved the best model. Now, how can I test model on new images (not validation)