Unable to reproduce the AoMSG-4 performance by provided checkpoint

[AkideLiu]

To clarify the discrepancy, we observed that the PP IoU values reported in Table 1 differ from those obtained during our local evaluation.

	PP IoU	PO IoU（GT）
paper	42.2	74.2
release model	39.54	67.11
our train model	31.89	50.88

[juexZZ]

Hi @AkideLiu,

Thank you for your interest!

Can you provide more details about your experiments, such as the evaluation thresholds, the backbone encoders, etc.? So that I can help check them on my end.

Feel free to shoot me an email :)

Best,
Juexiao