Inconsistency between reported metric and self-trained results

[seanzhuh]

Hi, I've traiend on waldo_kitchen of LERF dataset using your code but the reproduced result (Loc. Acc is 0.8182) was lower than that reported in your paper (0.955). What could be the reason?

[garrisonz]

More result for reference.

I train and eval on teatime of LERF dataset follow README instruction, using the default parameter.

	self-trained	in paper
mIoU scores	58.0	65.1
Localization accuracy	83.1	88.1

[AbdallahElsayed73]

I personally trained on Teatime dataset and I got 88.1 localization accuracy. The only difference I found with self-training was indeed the evaluation on Waldo Kitchen although I got 77.3% localization accuracy. I also got accuracy of 67.6% on ramen dataset while it was reported to be 73.2% in the paper