Could you specify which type of noise is this?
I don’t think it’s necessary to classify the type of noise in the seismic data, as the goal of the machine learning model is to remove blind noise. This is the predicted noise for a single slice from one cube in my current model. In fact, the model should be able to handle multiple types of noise.
all the models are overfitting somehow, I think noise levels are different in test dataset or maybe noise type. So that’s why I asked it could be gaussian, poison etc.
I’m confident that the training dataset includes the same noise types found in the test dataset since the submitted results have high SSIM scores, similar to the training dataset.
What is the difference between the SSIM of the training dataset and the test dataset?
0.94 eval and 0.59 test
Have you checked the submitted file names? The pre_filenames are as follows:
['2024-06-10_0d6402b1', '2024-06-10_1a4e5680', '2024-06-10_1b9a0096', '2024-06-10_2bd82c05', '2024-06-10_3b118e17', '2024-06-10_43537d46', '2024-06-10_662066f4', '2024-06-10_971ac6dd', '2024-06-10_9871c8c6', '2024-06-10_b7c329be', '2024-06-10_bfd43f22', '2024-06-10_c952ed24', '2024-06-10_cec3da7f', '2024-06-10_eb45f27e', '2024-06-11_f46c20fe']
The crossline and incline have the following nicknames:
data_keys_x = [pre_filenames[i] + "_gt.npy-x_0", pre_filenames[i] + "_gt.npy-x_1", pre_filenames[i] + "_gt.npy-x_2"]data_keys_i = [pre_filenames[i] + "_gt.npy-i_0", pre_filenames[i] + "_gt.npy-i_1", pre_filenames[i] + "_gt.npy-i_2"]
Don’t mess the crossline and incline.
If you submit the original noise file, you will get a score of 0.9017. ps: the results should be normalized to a 0-255 range.
file_names = [‘2024-06-10_0d6402b1’, ‘2024-06-10_1a4e5680’, ‘2024-06-10_1b9a0096’, ‘2024-06-10_2bd82c05’, ‘2024-06-10_3b118e17’, ‘2024-06-10_43537d46’, ‘2024-06-10_662066f4’, ‘2024-06-10_971ac6dd’, ‘2024-06-10_9871c8c6’, ‘2024-06-10_b7c329be’, ‘2024-06-10_bfd43f22’, ‘2024-06-10_c952ed24’, ‘2024-06-10_cec3da7f’, ‘2024-06-10_eb45f27e’, ‘2024-06-11_f46c20fe’], after this I did rescale_volumn then create submission functions, but I am not sure what’s wrong, the models outputs the same images, it doesn’t remove any noise.