How did you pre-process the image bands?

[o0vv0o]

Hello! I am a beginner in the field of deep learning and landslide detection. I am interested in the way you re-scale the image data. For example, the slope may range from 0-90 degrees, and the elevation may range from 0-4000 meters. Besides, the optical bands and DEM data also have various ranges. So how did you normalize these bands? Thank you!

[YonghaoXu]

Hi, thanks for your interest in the Landslide4Sense dataset.

The 14 bands of the data are simply divided by a mean vector to implement the normalization: img_mean = [1111.81236406, 824.63171476, 663.41636217, 445.17289745, 645.8582926, 1547.73508126, 1960.44401001, 1941.32229668, 674.07572865, 9.04787384, 1113.98338755, 519.90397929, 20.29228266, 772.83144788]

[jerry840712]

Hi

Thanks for preparing the dataset. Now I want to do the prediction on other more recent images. I would like to which level of Sentinel-2 images you used in this dataset, is it L1C or L2A? For the mean value from the list img_mean above, the img_mean[9] is 9.04787384, which is lower then other bands, could you help me to check the value is right?

Thanks. Chen-Wei

[Shan2L]

Hi，@YonghaoXu, Thank you for your preparation for the data. I have some problem for normalizing the data, too. The image data are in an uncertain range about 0 -2. I used the mean and std you supplied to normalize the data, but it doesn't work. I want to normalize the RGB channel to 0~1, how to do that? Looking for your reply.

Best Regrads, Shanlin

[fsqy-zhang]

嗨，@YonghaoXu，感谢您对数据的准备。我在规范化数据方面也遇到了一些问题。图像数据处于大约 0 -2 的不确定范围内。我使用了您提供的平均值和标准来规范化数据，但它不起作用。我想将RGB通道标准化为0~1，该怎么做？寻找您的回复。

最佳毕业生，山林

Hello, I have encountered a similar problem. My idea is to obtain the original image data, but the data value distribution obtained through h5 is about 0-33. How did you solve it? Thank you very much!

[fsqy-zhang]

嗨，感谢您对Landslide4Sense数据集的关注。

数据的 14 个波段简单地除以均值向量以实现归一化： img_mean = [1111.81236406, 824.63171476, 663.41636217, 445.17289745, 645.8582926, 1547.73508126, 1960.44401001, 1941.32229668, 674.07572865, 9.04787384, 1113.98338755, 519.90397929, 20.29228266, 772.83144788]

Thank you for providing the dataset. Did your standardization process directly divide by the values you provided? According to this value, it seems that it cannot be restored to true color images. I look forward to your reply

[wbuhuibiancheng]

嗨，感谢您对Landslide4Sense数据集的关注。 数据的 14 个波段简单地除以均值向量以实现归一化： img_mean = [1111.81236406, 824.63171476, 663.41636217, 445.17289745, 645.8582926, 1547.73508126, 1960.44401001, 1941.32229668, 674.07572865, 9.04787384, 1113.98338755, 519.90397929, 20.29228266, 772.83144788]

Thank you for providing the dataset. Did your standardization process directly divide by the values you provided? According to this value, it seems that it cannot be restored to true color images. I look forward to your reply

I also encountered this problem, have you solved it?

[keepfighting666]

Hi, thanks for your interest in the Landslide4Sense dataset.

The 14 bands of the data are simply divided by a mean vector to implement the normalization: img_mean = [1111.81236406, 824.63171476, 663.41636217, 445.17289745, 645.8582926, 1547.73508126, 1960.44401001, 1941.32229668, 674.07572865, 9.04787384, 1113.98338755, 519.90397929, 20.29228266, 772.83144788] Hi, I want to know how you get these values? Just use total pixel value/counts ? Looking forward to your reply.

[BNU-yb96]

嗨，感谢您对Landslide4Sense数据集的关注。 数据的 14 个波段简单地除以均值向量以实现归一化： img_mean = [1111.81236406, 824.63171476, 663.41636217, 445.17289745, 645.8582926, 1547.73508126, 1960.44401001, 1941.32229668, 674.07572865, 9.04787384, 1113.98338755, 519.90397929, 20.29228266, 772.83144788]

Thank you for providing the dataset. Did your standardization process directly divide by the values you provided? According to this value, it seems that it cannot be restored to true color images. I look forward to your reply

I also encountered this problem, have you solved it?

I suspect that this set of parameters is just the mean of 14 bands in a certain sample? Not applicable to all samples?

[BNU-yb96]

Hi, thanks for your interest in the Landslide4Sense dataset.

The 14 bands of the data are simply divided by a mean vector to implement the normalization: img_mean = [1111.81236406, 824.63171476, 663.41636217, 445.17289745, 645.8582926, 1547.73508126, 1960.44401001, 1941.32229668, 674.07572865, 9.04787384, 1113.98338755, 519.90397929, 20.29228266, 772.83144788]

May I ask if your mean vector belongs to all datasets? Or is it a training dataset? Or is it from a certain sample?