Reproducing Results from Table 2

[wuwuwuxxx]

Thank you for sharing your insightful paper and making the code publicly available. I've been working on reproducing the results from Table 2, but I'm encountering some discrepancies compared to the reported values.

Here are the results I obtained using the provided configurations:

{"rois.thing.macc1": 0.1817628860473633, "rois.thing.macc5": 0.33184242248535156, "rois.stuff.macc1": 0.13718507874686764, "rois.stuff.macc5": 0.36789228331367924, "crops.thing.macc1": 0.6843925476074219, "crops.thing.macc5": 0.8644012451171875, "crops.stuff.macc1": 0.22904190927181603, "crops.stuff.macc5": 0.5747173957104953, "maskpool.thing.macc1": 0.17424376010894777, "maskpool.thing.macc5": 0.32028732299804685, "maskpool.stuff.macc1": 0.16302821321307487, "maskpool.stuff.macc5": 0.40581411685583724}

I've carefully followed the instructions in the repository and used the same configurations, but my results consistently differ from those in the paper. Could you please advise on:

Any potential implementation details that might affect reproducibility?
Whether there are additional hyperparameters or environment settings that weren't explicitly mentioned?

I'd be grateful for any insights that could help resolve this difference. Thank you for your time and consideration!