About the overflow problem of SAR data

[try-agaaain]

This is an excellent dataset, and I am deeply grateful to the author for making it publicly available. It has been extremely helpful for my work. However, there is a issue that users should be aware of when using this dataset:

The values of SAR (Synthetic Aperture Radar) images typically range between 6500 and 6600, but since unsigned int can only represent numbers between 0 and 65535, excessively large values will overflow. This causes some of the values to wrap around and fall into the range of 0 to 100.

When visualizing SAR images, this overflow can result in certain areas appearing unusually dark. This is a clear sign of the overflow issue.

Fortunately, this issue is relatively straightforward to address. I resolved it by applying the following processing function to the SAR images:

def lower_upper_bound(x):
    # Overflow results in values wrapping around to the range of 0–100.
    if x.min() < 2000:  # Overflow detected
        x_t = np.zeros_like(x)
        x_t[x > 2000] = x[x > 2000] - x[x > 2000].min()  # Shift non-overflowed values downward
        x_t[x <= 2000] = x[x <= 2000] + x_t.max()        # Shift overflowed values upward
        return x_t
    return x

This issue may not significantly impact usage, but I think it’s worth highlighting here to ensure a smoother experience when working with this dataset.