search

youngwanLEE / centermask2

[CVPR 2020] CenterMask : Real-time Anchor-Free Instance Segmentation
Other
772 stars 159 forks source link

The gap between the results of our training centermask_R_101_FPN_ms_3x.yaml and the record of the paper #85

Closed luyao-cv closed 3 years ago

luyao-cv commented 3 years ago

We used the configuration of centermask_R_101_FPN_ms_3x.yaml and ran it, and found that the performance of the mask ap was only 38.7 on test-dev. But in the table 5 of your paper, this mask ap achieved at 39.8. Can you share me with your centermask_R_101_FPN_ms_3x weights?

youngwanLEE commented 3 years ago

@luyao-cv Hi,

I share the model weight, metric, train_log and test-dev result.

luyao-cv commented 3 years ago

@luyao-cv Hi,

I share the model weight, train log, and test-dev result.

thank you very much for your sharing, but the train log you provide is the metrics, we want to know why the gap is so big, could you please share your train log, we want to know whether the config we use is wrong.

youngwanLEE commented 3 years ago

@luyao-cv update above comment. check it.

youngwanLEE commented 3 years ago

Please re-open this issue if you face another problem.