Shape/vol mismatch between noise and seismic volumes

Image Impeccable
1

noise_vol - ./training_data/77692237/seismic_w_noise_vol_77692237.npy and its seismic volume have different shapes, 1259 300 300 , 300 300 1259 respectively, and it seems the vols dont visually match.

There are like 213 vols with shape mismatch

@discourse-admin assist

2

Thanks for flagging this to us, @moruridarius. The ThinkOnward team is investigating. When loading the seismic data please transpose the seismic volume like this: seismic = np.load('./seismic/data/path').T

This should resolve the issue.

ThinkOnward Team

4

@discourse-admin , how to make sure that transposed image is correct, i.e it is transposed according to the input image? because we have 300[0],300[1],1259[2], but when transposed, it can be transposed in two ways: 1259[2],300[0],300[1] and 1259[2],300[1],300[0].
Now how can I make sure which transpose matches with the actual input?

1 Like
6

You can visualize one sliced pair from each of the 250 paired 3D cubes. This approach can help identify and correct any mismatch issues quickly and efficiently.

7

Hello, in most cases, the issue can be resolved by changing the original shape from (d1, d2, d3) to the transposed shape (d3, d2, d1) using np.load().T. However, there are nine cubes that are completely mismatched: ["42673698", "77692226", "77692237", "77692243", "77692246", "77702634", "77702638", "89194322", "89194324"]. The noise cube and the clear label data for these cubes are from two different 3D cubes.