Result of refinement by SegFix on HRNet / HRNet-Semantic-Segmentation open source

[Jonnyboyyyy]

From my understanding, the two open source (HRNet-Semantic-Segmentation & openseg.pytorch) doesn't differ greatly.

So I applied SegFix to results generated from HRNet-Semantic-Segmentation. The original mIoU is like below.

Obviously, I assumed that the final mIoU after applying SegFix would increase. However, that's not the case. mIoU actually decreased to 80.29.

I applied SegFix the way described in MODEL_ZOO.md (below)

Is this the correct way to apply SegFix? Or is there any other way to apply SegFix?