what does" bm = (2 * (bm / 286.8) - 1) * 0.99" in inference.py mean?

[quynhanh12345]

[fh2019ustc]

Hi, we use the code "bm = (2 * (bm / 286.8) - 1) * 0.99" rather than "bm = (2 * (bm / 287) - 1)". The goal is to compensate the coordinate bias from the dataset, which would cause the shifting of the rectified image.

[lsabrinax]

did the code "bm = (2 * (bm / 286.8) - 1) * 0.99" use on the pred bm when calculate the BM L1 loss

[fh2019ustc]

@lsabrinax It is not used to calculate the flow loss.

[lsabrinax]

@lsabrinax It is not used to calculate the flow loss.

thank you for your reply, does it mean the GeoTr model directly regress the absolute value in flow between 0-287

[fh2019ustc]

@lsabrinax Yeah~