About part segmentation best inctance avg mIOU

[sh4175515]

In my experiment based on shapenet, thebest inctance avg mIOU is only 84.4%, which is 2% lower than that in the paper. Do you know the reason？Thank you very much

[lyb92nlf]

In my experiment based on shapenet, thebest inctance avg mIOU is only 84.4%, which is 2% lower than that in the paper. Do you know the reason？Thank you very much

can I see your hyper para?why i just have 82.9% mIoU?Thank you!

[qq456cvb]

A ~2% gap is about what I'd expect from this re-implementation. It isn't trying to exactly match the paper's pipeline (augmentation, voting, schedule all differ). Training longer plus a little tuning usually recovers part of it, but some gap is normal here.