bg_num_rois = 0 and fg_num_rois = 0, this should not happen!

yanxp / MetaR-CNN

Meta R-CNN : Towards General Solver for Instance-level Low-shot Learning

https://yanxp.github.io/metarcnn.html

177 stars 23 forks source link

bg_num_rois = 0 and fg_num_rois = 0, this should not happen! #57

Open yy995926 opened 3 years ago

yy995926 commented 3 years ago

1618454240(1) I always have the following problems in the first stage of training. I lowered the learning rate and could train later, but the results were very low. Does anyone know how to solve this problem