epe evaluation metric is different from RAFT-stereo

[Guptajakala]

Hi, I noticed that you are using L1 loss for computing EPE loss on sceneflow.

IGEV/IGEV-Stereo/evaluate_stereo.py

Line 143 in 60a0f9c

# epe = torch.sum((flow_pr - flow_gt)**2, dim=0).sqrt()

However, RAFT-stereo used sqrt of L2 loss
https://github.com/princeton-vl/RAFT-Stereo/blob/6068c1a26f84f8132de10f60b2bc0ce61568e085/evaluate_stereo.py#L130

These two metrics are not the same. Which one did you use to obtain the results in the paper?