Training configuration on KINS dataset

[trqminh]

Hi @YutingXiao,
Thank you for your consideration.

I have a question about reproducing results. Indeed, I am reproducing Mask RCNN performance on the KINS dataset. I follow this config file to train, but my results are lower than those reported in Table 1 of the paper (e.g. AP of maskrcnn: 27.91 of mine compare to 30.01 of yours). Did you use the same config as I am using here?

Below I add the screenshot of my results and yours so you can have a clearer look. Thank you!

[trqminh]

My bad. I forgot to change the pre-trained weight to R101 in the config file.